Skip to content
This repository has been archived by the owner on Oct 11, 2024. It is now read-only.

Commit

Permalink
switch to GCP based build VM (#201)
Browse files Browse the repository at this point in the history 
SUMMARY:
* switch over to GCP VM's for building stage of "remote push"

NOTE: this is just the start. i'll redo the benchmarking and nightly
workflows in an upcoming PR.

TEST PLAN:
runs on remote push

Co-authored-by: andy-neuma <[email protected]>
  • Loading branch information
@andy-neuma
andy-neuma and andy-neuma authored Apr 23, 2024
1 parent e8e00d2 commit df1f1a0
Show file tree
Hide file tree
Showing 2 changed files with 8 additions and 8 deletions.
2 changes: 1 addition & 1 deletion .github/workflows/build-test.yml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -69,7 +69,7 @@ jobs:
BUILD:
uses: ./.github/workflows/build.yml
with:
build_label: aws-avx512-192G-4-T4-64G
build_label: ${{ inputs.build_label }}
timeout: ${{ inputs.timeout }}
gitref: ${{ inputs.gitref }}
Gi_per_thread: ${{ inputs.Gi_per_thread }}
Expand Down
14 changes: 7 additions & 7 deletions .github/workflows/remote-push.yml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -18,11 +18,11 @@ jobs:
python: [3.10.12]
uses: ./.github/workflows/build-test.yml
with:
build_label: aws-avx512-192G-4-T4-64G
timeout: 360
build_label: gcp-build-static
timeout: 240
gitref: '${{ github.ref }}'
Gi_per_thread: 4
nvcc_threads: 8
Gi_per_thread: 1
nvcc_threads: 4
python: ${{ matrix.python }}
test_skip_list: neuralmagic/tests/skip-for-remote-push.txt
secrets: inherit
Expand All @@ -35,8 +35,8 @@ jobs:
benchmark_config_list_file: ./.github/data/nm_benchmark_remote_push_configs_list.txt
timeout: 180
gitref: '${{ github.ref }}'
Gi_per_thread: 12
nvcc_threads: 1
python: "3.10.12"
Gi_per_thread: 1
nvcc_threads: 4
python: 3.10.12
push_benchmark_results_to_gh_pages: "false"
secrets: inherit

13 comments on commit df1f1a0

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

bigger_is_better

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.288246490356173 prompts/s 13.304553986613938 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 863.7360218731512 tokens/s 864.796009129906 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.069935100521777 prompts/s 7.051461020667885 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 919.091563067831 tokens/s 916.689932686825 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.25629060431578965 prompts/s 0.2562593094265349 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 33.317778561052656 tokens/s 33.31371022544953 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.91264545176513 prompts/s 0.9127306469948456 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.6439087294669 tokens/s 118.65498410932993 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.372599423902794 prompts/s 3.3618719190890656 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1068.7430314405563 tokens/s 1065.343592440134 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 770.9155215145485 tokens/s 768.4678895051088 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - synthetic\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.8020361908600644 prompts/s 3.803884961395343 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - synthetic\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1459.9818972902647 tokens/s 1460.6918251758116 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.751349280304105 prompts/s 4.748476807201662 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 617.6754064395336 tokens/s 617.3019849362161 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.24027857686758167 prompts/s 0.2402768434229402 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.236214992785616 tokens/s 31.235989644982226 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.85063852636867 prompts/s 25.284467324256294 prompts/s 0.98
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1680.2915042139634 tokens/s 1643.490376076659 tokens/s 0.98
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 48.92551975017248 prompts/s 48.79581921111015 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3180.158783761211 tokens/s 3171.72824872216 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 8.022642464328776 prompts/s 8.014925201315489 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4115.615584200662 tokens/s 4111.656628274846 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.650050863429614 prompts/s 6.633908223317837 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 864.5066122458499 tokens/s 862.4080690313189 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.76449045559813 prompts/s 0.7639268340372671 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.3837592277569 tokens/s 99.31048842484472 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.939463859777623 prompts/s 1.938586235550944 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3973.9614486843498 tokens/s 3972.163196643884 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7928168522205612 prompts/s 1.7931727190644227 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.06619078867297 tokens/s 233.11245347837496 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.6489194857440275 prompts/s 6.652228260231785 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 864.3595331467236 tokens/s 864.7896738301321 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.2404804442982623 prompts/s 0.24030861859274258 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.262457758774097 tokens/s 31.240120417056538 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.6865156020312098 prompts/s 1.6806087990413068 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 550.0357010715201 tokens/s 548.1092720958677 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 383.33375290434714 tokens/s 381.9844535282326 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.9369899361293696 prompts/s 2.926136100906273 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 381.8086916968181 tokens/s 380.3976931178155 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 9.398768398550642 prompts/s 9.392961544017185 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2415.483478427515 tokens/s 2413.9911168124167 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9125586774860674 prompts/s 0.9123163639924358 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.63262807318877 tokens/s 118.60112731901664 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.8161797845745946 prompts/s 1.8186241599053863 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 577.056013739409 tokens/s 577.832666742845 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 406.6680802832358 tokens/s 407.20328597998866 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.470015691101717 prompts/s 3.47554536374714 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1099.613272353223 tokens/s 1101.365570317831 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 793.5440083352872 tokens/s 794.5305234247786 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.7259007442584813 prompts/s 2.7244396666690345 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 354.36709675360254 tokens/s 354.1771566669745 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.217212893910394 prompts/s 2.2167695683344157 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 684.7344673167006 tokens/s 684.5975565568232 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 508.64933185005424 tokens/s 508.8165968661773 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.465804543468793 prompts/s 7.4600015789981065 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3829.957730799491 tokens/s 3826.980810026029 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.9759590872115957 prompts/s 1.975317062191078 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 256.87468133750747 tokens/s 256.7912180848401 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.4506465240510393 prompts/s 2.4510298349474677 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 756.8249972009356 tokens/s 756.9433738273766 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 562.9853922058959 tokens/s 562.7597181437185 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.427405392446145 prompts/s 11.415719977955193 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1485.562701017999 tokens/s 1484.0435971341751 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.134317163132327 prompts/s 25.31894400415064 prompts/s 1.01
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1633.7306156036013 tokens/s 1645.7313602697916 tokens/s 1.01
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.491221400565894 prompts/s 3.4889400599782956 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3578.5019355800414 tokens/s 3576.163561477753 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.24032183084421932 prompts/s 0.24029023094156635 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.241838009748513 tokens/s 31.237730022403625 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.4365038924401476 prompts/s 3.43365128693395 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1088.9937184753583 tokens/s 1088.0897563164995 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 785.5412607625135 tokens/s 784.884613075084 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9839790793228471 prompts/s 0.9839491630042055 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.12951146860587 tokens/s 290.12069054233336 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.42140364421624 tokens/s 212.39854615663114 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.57013596941793 prompts/s 23.549919588363764 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3040.547540054913 tokens/s 3037.9396268989253 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0716590898225413 prompts/s 2.0683856916680297 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4244.829475046387 tokens/s 4238.122282227793 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.180574951230028 prompts/s 3.179495041356072 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1499.2148924614935 tokens/s 1498.7058596638462 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 19.67660464780448 prompts/s 19.663083685272863 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2557.9586042145825 tokens/s 2556.2008790854725 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.0800412918181 prompts/s 7.079761182632859 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3310.344106402471 tokens/s 3310.213138551819 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 28.850313839568255 prompts/s 28.794803850253647 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3721.690485304305 tokens/s 3714.5296966827204 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.46600721872350653 prompts/s 0.465998128153441 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 126.01456537241914 tokens/s 126.01210716106583 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.08312250974507 tokens/s 118.08081901989425 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.815321152307003 prompts/s 13.854177055679637 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3550.5375361428996 tokens/s 3560.523503309667 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.585295086815281 prompts/s 5.583551563685145 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2611.4605707913524 tokens/s 2610.645369116626 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.390216819401325 prompts/s 4.390290227894882 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1391.215807900086 tokens/s 1391.239070317609 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1006.1294029919053 tokens/s 1003.5588820335794 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.21452734631685 prompts/s 24.22708159873138 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3123.6740276748737 tokens/s 3125.293526236348 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.617908519372637 prompts/s 24.758579707575098 prompts/s 1.01
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1600.1640537592216 tokens/s 1609.3076809923814 tokens/s 1.01
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.279366750662627 prompts/s 6.262518754196035 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 816.3176775861415 tokens/s 814.1274380454846 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7940838152625345 prompts/s 1.7930718518766309 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.23089598412946 tokens/s 233.09934074396202 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.6966985711557236 prompts/s 3.6969976529082196 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 480.57081425024404 tokens/s 480.6096948780686 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.7620666359455757 prompts/s 3.752344427261337 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3856.118301844215 tokens/s 3846.1530379428705 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.78601484116294 prompts/s 5.780744121662275 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2968.225613516588 tokens/s 2965.5217344127473 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4740981087206661 prompts/s 0.47413571902199947 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.32309405922088 tokens/s 125.33303596627533 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.2445374255261 tokens/s 99.25241051527189 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.589916568452576 prompts/s 5.583649217567208 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2613.6213907456868 tokens/s 2610.691028165724 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.42641556594017 prompts/s 11.384394975067053 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1485.434023572222 tokens/s 1479.971346758717 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.15925613244758 prompts/s 10.14320166242732 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1320.7032972181853 tokens/s 1318.6162161155517 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9835880870407453 prompts/s 0.983631103681614 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.01422609092054 tokens/s 290.0269096908695 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.29765270687446 tokens/s 212.30693741863956 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.7615196539434037 prompts/s 3.7569492477931656 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3855.557645291989 tokens/s 3850.8729789879944 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.209500442218033 prompts/s 10.152973049137339 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1327.2350574883444 tokens/s 1319.886496387854 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.7748380126551354 prompts/s 0.7735157651280619 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 100.72894164516761 tokens/s 100.55704946664804 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4919100992277446 prompts/s 0.4918972664410856 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 130.031515629862 tokens/s 130.02812341103657 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.99698400914218 tokens/s 121.99052207738923 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0410845542069347 prompts/s 2.0384319394042425 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4182.18225157001 tokens/s 4176.747043839293 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.828019842159 prompts/s 25.81640908488198 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1678.8212897403353 tokens/s 1678.0665905173287 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.001101418232668 prompts/s 2.002675145423224 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4100.256805958737 tokens/s 4103.481372972186 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.041881096674683 prompts/s 2.037665339068763 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4183.814367086426 tokens/s 4175.176279751895 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 14.184913253045174 prompts/s 14.13566061315486 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3645.5227060326097 tokens/s 3632.864777580799 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.014534154666631 prompts/s 1.0135982483583388 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 131.88944010666205 tokens/s 131.76777228658406 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4918843882856642 prompts/s 0.4918873995766033 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 130.0247191994325 tokens/s 130.0255152040793 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.01028289963139 tokens/s 121.99791284298915 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.526130694053267 prompts/s 11.50371467682034 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1498.3969902269246 tokens/s 1495.4829079866442 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9680734653891591 prompts/s 0.9680309403748015 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 288.65691899818813 tokens/s 288.64423903115704 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 210.04289978548582 tokens/s 210.04012667272318 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 15.871073556042079 prompts/s 15.85819651376316 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4078.8659039028144 tokens/s 4075.5565040371316 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.18855735080431 prompts/s 24.194691196838754 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3120.323898253756 tokens/s 3121.115164392199 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9132682543350292 prompts/s 0.9127050864021167 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.7248730635538 tokens/s 118.65166123227517 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.032511576750037 prompts/s 6.036791955001028 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2820.561112825247 tokens/s 2822.5624464802804 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 8.201743131083298 prompts/s 8.19876334119261 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4207.494226245732 tokens/s 4205.965594031809 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0407406983382455 prompts/s 2.036458994186228 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4181.477690895065 tokens/s 4172.704479087582 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.633115132145545 prompts/s 6.631018479858359 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 862.304967178921 tokens/s 862.0324023815867 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.2912676938151155 prompts/s 2.290440902922561 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 707.6045643219427 tokens/s 707.3492292465648 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 458.9286989768006 tokens/s 458.4974060258344 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4640097572791887 prompts/s 0.4639583106048493 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.65633923918075 tokens/s 122.64273982528587 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 115.08679339876625 tokens/s 115.07403325161876 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.3747107938504493 prompts/s 3.372912386549949 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.7124032005584 tokens/s 438.47861025149336 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.903812893832965 prompts/s 13.862772772855124 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3573.2799137150723 tokens/s 3562.732602623767 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9475421758979542 prompts/s 0.9475222154869255 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 279.38596903743144 tokens/s 279.38008364370495 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 204.53961256325206 tokens/s 204.5353038423927 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.917713308320806 prompts/s 23.909559126416504 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3085.385016773384 tokens/s 3084.3331273077292 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - 2:4 Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.589919954647531 prompts/s 5.578648697507994 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - 2:4 Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2613.622973995 tokens/s 2608.3529850068376 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.584353882371954 prompts/s 5.581164920707232 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 725.966004708354 tokens/s 725.5514396919401 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.983892797392376 prompts/s 0.9838806054260277 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.10407095380003 tokens/s 290.10047611188236 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.46509031156765 tokens/s 212.40342470138896 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.467331135569104 prompts/s 7.459659949219531 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3830.7408725469504 tokens/s 3826.8055539496195 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.117041941446385 prompts/s 4.119903332561249 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1304.649420824945 tokens/s 1305.5561670553343 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 821.5255277747223 tokens/s 822.7144828880427 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12.24431932757904 prompts/s 12.264694778738873 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1579.5171932576961 tokens/s 1582.1456264573146 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 16.841597240811726 prompts/s 16.74529333683377 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2189.4076413055245 tokens/s 2176.8881337883904 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.865534536170695 prompts/s 3.8650348835012154 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3962.1728995749627 tokens/s 3961.660755588746 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.468505424500695 prompts/s 7.460202709235207 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3831.3432827688566 tokens/s 3827.083989837661 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.831217323530137 prompts/s 13.888661076110504 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3554.622852147245 tokens/s 3569.3858965603995 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.044098630880455 prompts/s 4.044106530230551 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4145.201096652467 tokens/s 4145.209193486315 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.3688370082571244 prompts/s 3.36660492732784 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 437.94881107342616 tokens/s 437.6586405526192 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.4288747664124415 prompts/s 6.3452258756993265 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 835.7537196336174 tokens/s 824.8793638409124 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.373190654738094 prompts/s 3.3672515848879416 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.51478511595224 tokens/s 437.7427060354324 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.3517787535966006 prompts/s 2.35158613748706 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 726.2919932107261 tokens/s 726.2325082196705 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 539.730088245415 tokens/s 539.6984248978303 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7956474696304574 prompts/s 1.7943553898000457 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.43417105195945 tokens/s 233.26620067400594 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.76270073290838 prompts/s 3.7554947864560835 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3856.7682512310894 tokens/s 3849.3821561174855 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.225320783752359 prompts/s 2.2244860054437767 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 687.2383999102951 tokens/s 686.9805981078501 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 510.8387049294348 tokens/s 510.5996197455427 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.353782021547051 prompts/s 6.344786048806903 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 825.9916628011166 tokens/s 824.8221863448973 tokens/s 1.00

This comment was automatically generated by workflow using github-action-benchmark.

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

smaller_is_better

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 65916.57909949754 ms 66787.84569050003 ms 0.99
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 22771.100657333376 ms 23200.36978493534 ms 0.98
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 17246.62806949891 ms 17407.502998499695 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 260.3458671893359 ms 261.4966492002461 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 278.48446449535345 ms 279.02716963159884 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 77236.82472850033 ms 78389.37180349967 ms 0.99
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 57611.70718832929 ms 58591.156928930686 ms 0.98
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 70195.4299455001 ms 72002.09738900048 ms 0.97
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 67.1830713443611 ms 68.32876132892734 ms 0.98
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 65.52266516921573 ms 65.84333552122904 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 258672.51598250004 ms 257960.56549199967 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 243841.84864052472 ms 243138.74433172468 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 245298.9168285012 ms 244561.7894119996 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 69.77612497205166 ms 70.2651706463995 ms 0.99
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 66.91512996603699 ms 66.51669527142957 ms 1.01
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 67226.0375980004 ms 66556.05931600257 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 29603.82816572403 ms 29395.456832029286 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 28459.95752499948 ms 28430.397764501322 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 188.3846817917835 ms 188.6404633420125 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 199.91143032979787 ms 200.07395705000116 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12292.36584499995 ms 12239.373073500246 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 200.15373098536412 ms 196.4258269800266 ms 1.02
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 173.80345649871742 ms 172.22716749984102 ms 1.01
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 95.75401380671806 ms 95.72504978387069 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 100.11628239847316 ms 100.80430680167527 ms 0.99
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2418.4398830002465 ms 2411.166190499898 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 115.76931008932297 ms 116.23203609204211 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 79.56178249969525 ms 80.91534550021606 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 18.101819158416294 ms 18.117534204784953 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 16.23795044677317 ms 16.20924310544079 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 63470.558906000406 ms 64089.48246949967 ms 0.99
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40796.949828582 ms 41101.79994840606 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40240.593307499694 ms 40744.35401400115 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.39904848222909 ms 99.1521024514054 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 101.50102836840136 ms 100.95686430257042 ms 1.01
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2134.7849715002667 ms 2124.492699500479 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 136.38976997335703 ms 136.58727046664958 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 104.76988149912358 ms 102.26584150041163 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 15.086482768662055 ms 15.11269000745712 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.666606953390275 ms 13.671886747409213 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6130.480080999405 ms 6157.483541000147 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 105.6716920200294 ms 105.50076459997702 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 65.50430149945896 ms 63.466036500358314 ms 1.03
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 34.67247563703323 ms 34.73502434048436 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 35.187566223780756 ms 35.23139902220215 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 14741.252231000544 ms 14700.193791500169 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 210.27452850930786 ms 208.177746926672 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 167.82330450041627 ms 164.7100544996647 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.75612429642501 ms 118.11084162295658 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.32679311695846 ms 125.06230301204924 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5214.891693500249 ms 5235.645166498216 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 92.3703036599424 ms 91.47958961346981 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 62.99134750224766 ms 61.74462950002635 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.19878991892422 ms 37.26582722049259 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.63833172599384 ms 31.708689124541355 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3655.2010885006894 ms 3651.734495499113 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 143.41733286998837 ms 143.39316453994493 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 92.92677350094891 ms 94.5331505008653 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.458502157872495 ms 24.4685511837163 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.753476690872578 ms 23.80589036230062 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2066.586602500138 ms 2067.7271800004746 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 116.29346694651758 ms 116.29355367333726 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 86.21472949926101 ms 86.71847000005073 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.641102608210412 ms 11.620019052349596 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.620520280604662 ms 11.619994200319992 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1956.6215115000887 ms 1925.8178315003533 ms 1.02
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 79.53619423332081 ms 79.36294155328748 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.98137199980556 ms 37.76919200026896 ms 1.09
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.014972965407692 ms 10.98752899201461 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.11630137046384 ms 11.090169800305919 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6053.694493500188 ms 6079.410813499635 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 120.83212203337703 ms 122.25182649330236 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 79.50092850023793 ms 83.33002849940385 ms 0.95
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.84340517129665 ms 39.852174955426804 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.449177643674645 ms 39.52926163180674 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6053.720715499367 ms 6064.388077998956 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.73355978801555 ms 122.99509105597099 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 92.58800700081338 ms 91.56941200126312 ms 1.01
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 55.06918555292367 ms 55.034764851352534 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 48.01629185927375 ms 48.14404020383918 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6383.334336999724 ms 6367.810279999503 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 111.97803860662438 ms 111.16892662006042 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 71.34634100020776 ms 69.513698499577 ms 1.03
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 36.284039144000666 ms 36.28406362460848 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 36.94600637013761 ms 36.9687584896116 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6055.742755500432 ms 6061.940364000293 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 128.9584407933156 ms 128.1418428899633 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 91.93805599988991 ms 87.2189094998248 ms 1.05
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.289583901993524 ms 40.458295919317536 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.36464489005277 ms 40.42338318231981 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1812.500326999725 ms 1813.6266555002294 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 92.7654148466839 ms 93.09651339993555 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 58.47218449889624 ms 60.07437150037731 ms 0.97
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12.54018810517509 ms 12.492343204249421 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.773172392356791 ms 11.77166285193115 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 18595.579219001593 ms 18630.98225700014 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1087.1886521253982 ms 1066.207715544699 ms 1.02
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 236.42433899840398 ms 232.56485849924502 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 161.72739512669037 ms 162.23560618336472 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 148.92583269572387 ms 149.11634172076177 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5381.206855501659 ms 5385.054345500976 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 182.7010252038987 ms 183.15599952013872 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 170.11806400114438 ms 174.65145149981254 ms 0.97
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.335607190915894 ms 40.29276032413655 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.612841919993144 ms 37.395030997344755 ms 1.01
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7864.310286499858 ms 7866.947419000098 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 156.9076700013587 ms 156.33848048406313 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 128.9679054998487 ms 125.77558449993376 ms 1.03
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 58.064269891361505 ms 57.926682457080105 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 57.69306854901845 ms 57.515965234391565 ms 1.00

This comment was automatically generated by workflow using github-action-benchmark.

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

bigger_is_better

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.20006972025561 prompts/s 24.22708159873138 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3121.808993912973 tokens/s 3125.293526236348 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.39058212586998 prompts/s 4.390290227894882 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1391.331569866938 tokens/s 1391.239070317609 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1003.8773325326652 tokens/s 1003.5588820335794 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.0747040660417815 prompts/s 7.079761182632859 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3307.8486331184954 tokens/s 3310.213138551819 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 19.723977214120207 prompts/s 19.663083685272863 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2564.117037835627 tokens/s 2556.2008790854725 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9136051753241461 prompts/s 0.9127050864021167 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.76867279213899 tokens/s 118.65166123227517 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - 2:4 Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.587268061786145 prompts/s 5.578648697507994 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - 2:4 Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2612.3830549687295 tokens/s 2608.3529850068376 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.266885359435994 prompts/s 6.262518754196035 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 814.6950967266793 tokens/s 814.1274380454846 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 15.818910428551627 prompts/s 15.85819651376316 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4065.459980137768 tokens/s 4075.5565040371316 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.6368881731094245 prompts/s 6.633908223317837 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 862.7954625042252 tokens/s 862.4080690313189 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.6843017080910048 prompts/s 1.6806087990413068 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 549.3136676055788 tokens/s 548.1092720958677 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 382.8395325134746 tokens/s 381.9844535282326 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9836461965707789 prompts/s 0.983631103681614 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.03135987954937 tokens/s 290.0269096908695 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.31019506783693 tokens/s 212.30693741863956 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.4752813611412257 prompts/s 3.47554536374714 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1101.281910532043 tokens/s 1101.365570317831 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 794.7041731233678 tokens/s 794.5305234247786 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.179288697213479 prompts/s 3.179495041356072 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1498.6085960507287 tokens/s 1498.7058596638462 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 9.388267000003504 prompts/s 9.392961544017185 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2412.784619000901 tokens/s 2413.9911168124167 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.148025066687639 prompts/s 10.14320166242732 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1319.243258669393 tokens/s 1318.6162161155517 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.24037272040186206 prompts/s 0.2402768434229402 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.248453652242066 tokens/s 31.235989644982226 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.24036715461751065 prompts/s 0.24030861859274258 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.247730100276385 tokens/s 31.240120417056538 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.0142780566093703 prompts/s 1.0135982483583388 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 131.85614735921814 tokens/s 131.76777228658406 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4740786439461014 prompts/s 0.47413571902199947 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.31794874071244 tokens/s 125.33303596627533 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.24046279938389 tokens/s 99.25241051527189 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.7749154986611321 prompts/s 0.7735157651280619 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 100.73901482594717 tokens/s 100.55704946664804 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9839303442357208 prompts/s 0.9839491630042055 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.1151417657164 tokens/s 290.12069054233336 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.39776364235092 tokens/s 212.39854615663114 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.4439894391094463 prompts/s 3.43365128693395 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1091.3658133593924 tokens/s 1088.0897563164995 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 787.29368978746 tokens/s 784.884613075084 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.46592955080932486 prompts/s 0.465998128153441 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.99356293285223 tokens/s 126.01210716106583 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.06344197807752 tokens/s 118.08081901989425 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.403600274723948 prompts/s 11.384394975067053 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1482.4680357141133 tokens/s 1479.971346758717 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.7832111148542 prompts/s 5.780744121662275 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2966.7873019202043 tokens/s 2965.5217344127473 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.3713844150690897 prompts/s 3.36660492732784 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.27997395898166 tokens/s 437.6586405526192 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.9767747333074825 prompts/s 1.975317062191078 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 256.9807153299727 tokens/s 256.7912180848401 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.919601245272034 prompts/s 23.909559126416504 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3085.6285606400925 tokens/s 3084.3331273077292 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 28.765769286977637 prompts/s 28.794803850253647 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3710.784238020115 tokens/s 3714.5296966827204 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.040779817400593 prompts/s 2.037665339068763 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4181.557845853815 tokens/s 4175.176279751895 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.84129100715639 prompts/s 25.81640908488198 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1679.6839154651652 tokens/s 1678.0665905173287 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.730868274717282 prompts/s 2.7244396666690345 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 355.0128757132466 tokens/s 354.1771566669745 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.451018286341343 prompts/s 2.4510298349474677 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 756.9398073098424 tokens/s 756.9433738273766 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 562.9335398849706 tokens/s 562.7597181437185 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.120376335624357 prompts/s 4.119903332561249 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1305.7060569960026 tokens/s 1305.5561670553343 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 823.8912236485469 tokens/s 822.7144828880427 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.7596232256185482 prompts/s 3.7569492477931656 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3853.613806259012 tokens/s 3850.8729789879944 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.695863778027903 prompts/s 23.549919588363764 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3056.7664273655996 tokens/s 3037.9396268989253 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.587481978034053 prompts/s 5.583649217567208 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2612.483073649602 tokens/s 2610.691028165724 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0018807549557645 prompts/s 2.002675145423224 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4101.8536669043615 tokens/s 4103.481372972186 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.748771198462615 prompts/s 4.748476807201662 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 617.34025580014 tokens/s 617.3019849362161 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.294744716527315 prompts/s 13.304553986613938 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 864.1584065742755 tokens/s 864.796009129906 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.7590538079713482 prompts/s 3.7554947864560835 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3853.030153170632 tokens/s 3849.3821561174855 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0694139892228027 prompts/s 2.0683856916680297 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4240.229263917523 tokens/s 4238.122282227793 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9839368332214459 prompts/s 0.9838806054260277 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.1170550647874 tokens/s 290.10047611188236 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.39588460862538 tokens/s 212.40342470138896 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.912528941517019 prompts/s 0.9123163639924358 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.62876239721247 tokens/s 118.60112731901664 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.7645721528818069 prompts/s 0.7639268340372671 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.3943798746349 tokens/s 99.31048842484472 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.938063964249544 prompts/s 1.938586235550944 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3971.093062747316 tokens/s 3972.163196643884 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.636887386708015 prompts/s 24.758579707575098 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1601.3976801360209 tokens/s 1609.3076809923814 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.405456499052091 prompts/s 11.415719977955193 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1482.7093448767719 tokens/s 1484.0435971341751 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - synthetic\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.801909943195392 prompts/s 3.803884961395343 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - synthetic\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1459.9334181870306 tokens/s 1460.6918251758116 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.455893024602395 prompts/s 7.4600015789981065 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3824.873121621029 tokens/s 3826.980810026029 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.6946652550798604 prompts/s 3.6969976529082196 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 480.30648316038184 tokens/s 480.6096948780686 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.81827348102006 prompts/s 13.888661076110504 prompts/s 1.01
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3551.2962846221553 tokens/s 3569.3858965603995 tokens/s 1.01
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.352947835904284 prompts/s 6.344786048806903 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 825.8832186675569 tokens/s 824.8221863448973 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.587751690221 prompts/s 5.583551563685145 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2612.609180279731 tokens/s 2610.645369116626 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7922500254739204 prompts/s 1.7931727190644227 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 232.99250331160965 tokens/s 233.11245347837496 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.24049136921444103 prompts/s 0.24029023094156635 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.263877997877334 tokens/s 31.237730022403625 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.2914056114921615 prompts/s 2.290440902922561 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 707.6471569784192 tokens/s 707.3492292465648 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 458.89521900231216 tokens/s 458.4974060258344 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.847781365615841 prompts/s 13.862772772855124 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3558.8798109632708 tokens/s 3562.732602623767 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.3730444093302125 prompts/s 3.372912386549949 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.49577321292765 tokens/s 438.47861025149336 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.161771199413371 prompts/s 10.152973049137339 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1321.0302559237382 tokens/s 1319.886496387854 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.360797963112455 prompts/s 3.3618719190890656 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1065.003266530706 tokens/s 1065.343592440134 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 768.2268821320729 tokens/s 768.4678895051088 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 48.796912129616594 prompts/s 48.79581921111015 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3171.7992884250784 tokens/s 3171.72824872216 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.457388704607667 prompts/s 7.459659949219531 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3825.6404054637333 tokens/s 3826.8055539496195 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.75813114421745 prompts/s 3.752344427261337 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3852.084422822886 tokens/s 3846.1530379428705 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.813996908983725 prompts/s 1.8186241599053863 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 576.3624472226716 tokens/s 577.832666742845 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 406.1635825716372 tokens/s 407.20328597998866 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.528415504677007 prompts/s 11.50371467682034 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1498.694015608011 tokens/s 1495.4829079866442 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.211463037884652 prompts/s 25.284467324256294 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1638.7450974625024 tokens/s 1643.490376076659 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.2166533794747987 prompts/s 2.2167695683344157 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 684.5616743386038 tokens/s 684.5975565568232 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 508.88450517062773 tokens/s 508.8165968661773 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0395373201986557 prompts/s 2.036458994186228 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4179.011969087046 tokens/s 4172.704479087582 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.351170454856403 prompts/s 2.35158613748706 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 726.1041343384534 tokens/s 726.2325082196705 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 539.6092938592435 tokens/s 539.6984248978303 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.57499748820536 prompts/s 5.581164920707232 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 724.7496734666969 tokens/s 725.5514396919401 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.457901210380264 prompts/s 7.460202709235207 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3825.9033209250756 tokens/s 3827.083989837661 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7939638179198527 prompts/s 1.7930718518766309 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.21529632958087 tokens/s 233.09934074396202 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 14.128635546953467 prompts/s 14.13566061315486 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3631.0593355670408 tokens/s 3632.864777580799 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.256406459753778 prompts/s 0.2562593094265349 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 33.332839767991146 tokens/s 33.31371022544953 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.034733389474113 prompts/s 6.036791955001028 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2821.599943582516 tokens/s 2822.5624464802804 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.3942490354355455 prompts/s 6.3452258756993265 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 831.252374606621 tokens/s 824.8793638409124 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.038207134391362 prompts/s 2.0384319394042425 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4176.2864183679 tokens/s 4176.747043839293 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9132343008510725 prompts/s 0.9127306469948456 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.72045911063941 tokens/s 118.65498410932993 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.044466596786675 prompts/s 4.044106530230551 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4145.578261706342 tokens/s 4145.209193486315 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9472492792240369 prompts/s 0.9475222154869255 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 279.2996074768047 tokens/s 279.38008364370495 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 204.47638691089801 tokens/s 204.5353038423927 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4918899469119448 prompts/s 0.4918873995766033 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 130.0261885667035 tokens/s 130.0255152040793 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.9231215079074 tokens/s 121.99791284298915 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.030260300855865 prompts/s 7.051461020667885 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 913.9338391112624 tokens/s 916.689932686825 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12.268967369706536 prompts/s 12.264694778738873 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1582.6967906921432 tokens/s 1582.1456264573146 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.9353099601034867 prompts/s 2.926136100906273 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 381.5902948134533 tokens/s 380.3976931178155 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.86309754053098 prompts/s 3.8650348835012154 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3959.6749790442545 tokens/s 3961.660755588746 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.374109086662627 prompts/s 3.3672515848879416 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.6341812661415 tokens/s 437.7427060354324 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.6416883298404175 prompts/s 6.652228260231785 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 863.4194828792542 tokens/s 864.7896738301321 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 8.20163684811053 prompts/s 8.19876334119261 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4207.439703080702 tokens/s 4205.965594031809 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.49131303706324 prompts/s 3.4889400599782956 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3578.595862989821 tokens/s 3576.163561477753 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.464006688252242 prompts/s 0.4639583106048493 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.65552797259765 tokens/s 122.64273982528587 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 115.08603219824275 tokens/s 115.07403325161876 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4918945098955709 prompts/s 0.4918972664410856 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 130.0273947457952 tokens/s 130.02812341103657 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.07510016915015 tokens/s 121.99052207738923 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.876368356790872 prompts/s 13.854177055679637 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3566.226667695254 tokens/s 3560.523503309667 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.163630304524446 prompts/s 24.194691196838754 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3117.1083092836534 tokens/s 3121.115164392199 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7948478517727906 prompts/s 1.7943553898000457 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.33022073046277 tokens/s 233.26620067400594 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 16.78621659751626 prompts/s 16.74529333683377 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2182.2081576771134 tokens/s 2176.8881337883904 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.253922509825237 prompts/s 25.31894400415064 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1641.5049631386405 tokens/s 1645.7313602697916 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.2254082316359503 prompts/s 2.2244860054437767 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 687.2654061486917 tokens/s 686.9805981078501 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 510.84987759947114 tokens/s 510.5996197455427 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9680426607906103 prompts/s 0.9680309403748015 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 288.6477337856749 tokens/s 288.64423903115704 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 210.03944292060802 tokens/s 210.04012667272318 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 8.020069164486243 prompts/s 8.014925201315489 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4114.295481381442 tokens/s 4111.656628274846 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.633521030903454 prompts/s 6.631018479858359 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 862.357734017449 tokens/s 862.0324023815867 tokens/s 1.00

This comment was automatically generated by workflow using github-action-benchmark.

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

smaller_is_better

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 14684.293790000993 ms 14700.193791500169 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 208.3220699653078 ms 208.177746926672 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 163.52807249950274 ms 164.7100544996647 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.27909656845316 ms 118.11084162295658 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 124.47824077741187 ms 125.06230301204924 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 78676.20362649995 ms 78389.37180349967 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 58531.132082698656 ms 58591.156928930686 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 71954.00605649957 ms 72002.09738900048 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 68.14038796325349 ms 68.32876132892734 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 64.92595071329512 ms 65.84333552122904 ms 0.99
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3681.198673499239 ms 3651.734495499113 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 143.5252729934291 ms 143.39316453994493 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 92.9505505009729 ms 94.5331505008653 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.473252835140016 ms 24.4685511837163 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.76110547244236 ms 23.80589036230062 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 67226.44485249839 ms 66556.05931600257 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 29609.394949770016 ms 29395.456832029286 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 28935.302789999696 ms 28430.397764501322 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 189.77414023760403 ms 188.6404633420125 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 200.35119987480914 ms 200.07395705000116 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5317.406906498945 ms 5235.645166498216 ms 1.02
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 97.22629859997444 ms 91.47958961346981 ms 1.06
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 72.09632149897516 ms 61.74462950002635 ms 1.17
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.79556657487593 ms 37.26582722049259 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 32.08784486182563 ms 31.708689124541355 ms 1.01
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2136.1864145001164 ms 2124.492699500479 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 136.5205156833872 ms 136.58727046664958 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 103.44628050097526 ms 102.26584150041163 ms 1.01
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 15.112199620192566 ms 15.11269000745712 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.701449799449973 ms 13.671886747409213 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 62772.602527000345 ms 64089.48246949967 ms 0.98
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40312.67313304599 ms 41101.79994840606 ms 0.98
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39637.395538999954 ms 40744.35401400115 ms 0.97
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 98.81408293959342 ms 99.1521024514054 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 100.97947547578033 ms 100.95686430257042 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6139.295400500487 ms 6157.483541000147 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 106.25955792006911 ms 105.50076459997702 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 69.21571899965784 ms 63.466036500358314 ms 1.09
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 34.70233731521278 ms 34.73502434048436 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 35.21299997463882 ms 35.23139902220215 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2448.5039289993438 ms 2411.166190499898 ms 1.02
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 116.79448861999845 ms 116.23203609204211 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 82.21035350015882 ms 80.91534550021606 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 17.922777405924435 ms 18.117534204784953 ms 0.99
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 16.200238407852396 ms 16.20924310544079 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 18313.699185999212 ms 18630.98225700014 ms 0.98
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1011.0077165439482 ms 1066.207715544699 ms 0.95
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 228.9288289994147 ms 232.56485849924502 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 161.50510215724378 ms 162.23560618336472 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 149.1556797529473 ms 149.11634172076177 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1816.4417999996658 ms 1813.6266555002294 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 93.24310573667996 ms 93.09651339993555 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 58.024013999784074 ms 60.07437150037731 ms 0.97
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12.466045600511196 ms 12.492343204249421 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.803352483970196 ms 11.77166285193115 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6062.616990000606 ms 6064.388077998956 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.15372097194874 ms 122.99509105597099 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 89.19384000000719 ms 91.56941200126312 ms 0.97
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 55.04163031944853 ms 55.034764851352534 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 48.08009942613016 ms 48.14404020383918 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 66824.92080950033 ms 66787.84569050003 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23289.607439991953 ms 23200.36978493534 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 17678.39180450028 ms 17407.502998499695 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 261.2510039341174 ms 261.4966492002461 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 278.3666064952644 ms 279.02716963159884 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 260040.2680320003 ms 257960.56549199967 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 244739.1784466027 ms 243138.74433172468 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 246372.5062650001 ms 244561.7894119996 ms 1.01
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 70.76599359391062 ms 70.2651706463995 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 66.7802752570491 ms 66.51669527142957 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12370.23476899958 ms 12239.373073500246 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 197.18242574671845 ms 196.4258269800266 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 168.20268300125463 ms 172.22716749984102 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 96.474893671509 ms 95.72504978387069 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 102.3616917322235 ms 100.80430680167527 ms 1.02
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5452.829542000472 ms 5385.054345500976 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 182.73248830001586 ms 183.15599952013872 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 175.9249924980395 ms 174.65145149981254 ms 1.01
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.55267440456968 ms 40.29276032413655 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.65844869364191 ms 37.395030997344755 ms 1.01
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6103.5300604999065 ms 6061.940364000293 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 126.7067479067858 ms 128.1418428899633 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 84.51540699934412 ms 87.2189094998248 ms 0.97
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.656371456455744 ms 40.458295919317536 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.40241560150337 ms 40.42338318231981 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1919.0515755008164 ms 1925.8178315003533 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 79.28845361332303 ms 79.36294155328748 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 38.64191800039407 ms 37.76919200026896 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.00241749943221 ms 10.98752899201461 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.097576991533218 ms 11.090169800305919 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6365.792071499527 ms 6367.810279999503 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 109.22379640664683 ms 111.16892662006042 ms 0.98
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 68.86481349920359 ms 69.513698499577 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 36.34777783012724 ms 36.28406362460848 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.01667508754563 ms 36.9687584896116 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2065.5033054999876 ms 2067.7271800004746 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 116.66689061995082 ms 116.29355367333726 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 86.45775749937457 ms 86.71847000005073 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.582812428876155 ms 11.620019052349596 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.576612422942894 ms 11.619994200319992 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7846.566335500029 ms 7866.947419000098 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 156.7358107093678 ms 156.33848048406313 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 127.33971199941152 ms 125.77558449993376 ms 1.01
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 58.13160607264793 ms 57.926682457080105 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 57.70580534053015 ms 57.515965234391565 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6072.354036499746 ms 6079.410813499635 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.10033801673126 ms 122.25182649330236 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 85.18675849973079 ms 83.33002849940385 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.86187873030412 ms 39.852174955426804 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.59639743342392 ms 39.52926163180674 ms 1.00

This comment was automatically generated by workflow using github-action-benchmark.

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'smaller_is_better'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.10.

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 72.09632149897516 ms 61.74462950002635 ms 1.17

This comment was automatically generated by workflow using github-action-benchmark.

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

bigger_is_better

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7949995753116363 prompts/s 1.7930718518766309 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.34994479051275 tokens/s 233.09934074396202 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.393030217397234 prompts/s 4.390290227894882 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1392.1073455910096 tokens/s 1391.239070317609 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1004.1002733564279 tokens/s 1003.5588820335794 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9133889047985051 prompts/s 0.9127050864021167 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.74055762380566 tokens/s 118.65166123227517 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.6946174199026083 prompts/s 3.6969976529082196 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 480.3002645873391 tokens/s 480.6096948780686 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.490781812200338 prompts/s 3.4889400599782956 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3578.051357505346 tokens/s 3576.163561477753 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.2405263171280148 prompts/s 0.2402768434229402 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.268421226641927 tokens/s 31.235989644982226 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.24042031982477546 prompts/s 0.24029023094156635 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.25464157722081 tokens/s 31.237730022403625 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.46394746364488987 prompts/s 0.4639583106048493 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.63987253989019 tokens/s 122.64273982528587 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 115.07134291629656 tokens/s 115.07403325161876 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.849880507966269 prompts/s 13.888661076110504 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3559.419290547331 tokens/s 3569.3858965603995 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 28.7134362033417 prompts/s 28.794803850253647 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3704.0332702310793 tokens/s 3714.5296966827204 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.272809491262083 prompts/s 6.262518754196035 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 815.4652338640709 tokens/s 814.1274380454846 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.9381093952561521 prompts/s 1.938586235550944 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3971.1861508798556 tokens/s 3972.163196643884 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4919096725477962 prompts/s 0.4918873995766033 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 130.03140284128443 tokens/s 130.0255152040793 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.99359879185346 tokens/s 121.99791284298915 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 19.820842796272125 prompts/s 19.663083685272863 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2576.7095635153764 tokens/s 2556.2008790854725 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9838938052573649 prompts/s 0.9839491630042055 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.10436812615154 tokens/s 290.12069054233336 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.38987609488984 tokens/s 212.39854615663114 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.8638644634079493 prompts/s 3.8650348835012154 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3960.461074993148 tokens/s 3961.660755588746 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.117409572072783 prompts/s 4.119903332561249 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1304.7659192941442 tokens/s 1305.5561670553343 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 822.2329668443612 tokens/s 822.7144828880427 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4659137686166886 prompts/s 0.465998128153441 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.98929521753416 tokens/s 126.01210716106583 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.05944287567812 tokens/s 118.08081901989425 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.459146812465956 prompts/s 7.459659949219531 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3826.542314795036 tokens/s 3826.8055539496195 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.7631667701621914 prompts/s 3.7554947864560835 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3857.245939416246 tokens/s 3849.3821561174855 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.180179521393914 prompts/s 3.179495041356072 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1499.0285002813637 tokens/s 1498.7058596638462 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.486001939890945 prompts/s 3.47554536374714 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1104.6791547320415 tokens/s 1101.365570317831 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 796.9604674926948 tokens/s 794.5305234247786 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.291278185782584 prompts/s 2.290440902922561 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 707.6078045212828 tokens/s 707.3492292465648 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 458.9582958001568 tokens/s 458.4974060258344 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.4088570019205005 prompts/s 6.3452258756993265 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 833.1514102496651 tokens/s 824.8793638409124 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12.260728962114257 prompts/s 12.264694778738873 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1581.6340361127393 tokens/s 1582.1456264573146 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.2563238123877657 prompts/s 0.2562593094265349 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 33.32209561040954 tokens/s 33.31371022544953 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.8938658129645458 prompts/s 0.9123163639924358 prompts/s 1.02
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 116.20255568539096 tokens/s 118.60112731901664 tokens/s 1.02
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.6825898623122404 prompts/s 1.6806087990413068 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 548.7553707882146 tokens/s 548.1092720958677 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 382.43697153152397 tokens/s 381.9844535282326 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.973973546673259 prompts/s 1.975317062191078 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 256.6165610675237 tokens/s 256.7912180848401 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.038172346968543 prompts/s 2.037665339068763 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4176.215138938544 tokens/s 4175.176279751895 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.3646494240684777 prompts/s 3.3618719190890656 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1066.22375599306 tokens/s 1065.343592440134 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 769.1162394493491 tokens/s 768.4678895051088 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.466646062767903 prompts/s 7.4600015789981065 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3830.389430199934 tokens/s 3826.980810026029 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.042907336464067 prompts/s 6.036791955001028 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2825.421754237139 tokens/s 2822.5624464802804 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.18519591304833 prompts/s 24.194691196838754 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3119.8902727832346 tokens/s 3121.115164392199 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.512743718391192 prompts/s 25.284467324256294 prompts/s 1.08
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1528.3283416954275 tokens/s 1643.490376076659 tokens/s 1.08
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.24036227983532377 prompts/s 0.24030861859274258 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.24709637859209 tokens/s 31.240120417056538 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 8.20449338757999 prompts/s 8.19876334119261 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4208.905107828536 tokens/s 4205.965594031809 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.63674894766033 prompts/s 6.631018479858359 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 862.7773631958428 tokens/s 862.0324023815867 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.646122492661698 prompts/s 6.652228260231785 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 863.9959240460207 tokens/s 864.7896738301321 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.7633501335405812 prompts/s 0.7639268340372671 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.23551736027557 tokens/s 99.31048842484472 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.414057007941404 prompts/s 11.415719977955193 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1483.8274110323823 tokens/s 1484.0435971341751 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 8.019730075902089 prompts/s 8.014925201315489 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4114.1215289377715 tokens/s 4111.656628274846 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0006474215098162 prompts/s 2.002675145423224 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4099.326566673613 tokens/s 4103.481372972186 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.855333771786881 prompts/s 13.854177055679637 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3560.8207793492284 tokens/s 3560.523503309667 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.794816340225349 prompts/s 1.7943553898000457 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.32612422929537 tokens/s 233.26620067400594 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.761073573796811 prompts/s 3.7569492477931656 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3855.1004131417317 tokens/s 3850.8729789879944 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.0137874502703725 prompts/s 1.0135982483583388 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 131.79236853514843 tokens/s 131.76777228658406 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4740013847384472 prompts/s 0.47413571902199947 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.29752604176113 tokens/s 125.33303596627533 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.22428987191495 tokens/s 99.25241051527189 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.6459456507085815 prompts/s 6.633908223317837 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 863.9729345921155 tokens/s 862.4080690313189 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.2239646879065154 prompts/s 2.2244860054437767 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 686.8196013505427 tokens/s 686.9805981078501 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 510.51257675206585 tokens/s 510.5996197455427 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.761579958043578 prompts/s 3.752344427261337 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3855.6194569946674 tokens/s 3846.1530379428705 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0715838921290795 prompts/s 2.0683856916680297 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4244.675394972484 tokens/s 4238.122282227793 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.351265085707312 prompts/s 2.35158613748706 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 726.1333588687036 tokens/s 726.2325082196705 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 539.5808519485711 tokens/s 539.6984248978303 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.293176419655584 prompts/s 13.304553986613938 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 864.0564672776129 tokens/s 864.796009129906 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0413029231642428 prompts/s 2.036458994186228 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4182.629689563533 tokens/s 4172.704479087582 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.2169341932834823 prompts/s 2.2167695683344157 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 684.6483971310936 tokens/s 684.5975565568232 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 508.5794835005194 tokens/s 508.8165968661773 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.587629320312933 prompts/s 5.583551563685145 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2612.5519650055153 tokens/s 2610.645369116626 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9680832064492432 prompts/s 0.9680309403748015 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 288.6598235550138 tokens/s 288.64423903115704 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 210.0514671913353 tokens/s 210.04012667272318 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.836820165944927 prompts/s 13.862772772855124 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3556.0627826478462 tokens/s 3562.732602623767 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.586044120030053 prompts/s 5.583649217567208 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2611.810788761252 tokens/s 2610.691028165724 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.7752212016181141 prompts/s 0.7735157651280619 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 100.77875621035484 tokens/s 100.55704946664804 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.875616061847403 prompts/s 23.909559126416504 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3079.954471978315 tokens/s 3084.3331273077292 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9474708026731057 prompts/s 0.9475222154869255 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 279.36492440417413 tokens/s 279.38008364370495 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 204.5273639370344 tokens/s 204.5353038423927 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9839586699454171 prompts/s 0.9838806054260277 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.1234936956394 tokens/s 290.10047611188236 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.4005983566509 tokens/s 212.40342470138896 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - synthetic\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.803037199516326 prompts/s 3.803884961395343 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - synthetic\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1460.366284614269 tokens/s 1460.6918251758116 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.47030287538148 prompts/s 7.460202709235207 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3832.2653750706995 tokens/s 3827.083989837661 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.021724243809558 prompts/s 23.549919588363764 prompts/s 1.02
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2969.802427451433 tokens/s 3037.9396268989253 tokens/s 1.02
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.0245452285002585 prompts/s 7.051461020667885 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 913.1908797050336 tokens/s 916.689932686825 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.581963705688526 prompts/s 5.581164920707232 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 725.6552817395084 tokens/s 725.5514396919401 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9131930530203444 prompts/s 0.9127306469948456 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.71509689264477 tokens/s 118.65498410932993 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 16.772815693353547 prompts/s 16.74529333683377 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2180.466040135961 tokens/s 2176.8881337883904 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7930686540576055 prompts/s 1.7931727190644227 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.09892502748872 tokens/s 233.11245347837496 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.4342641182384384 prompts/s 3.43365128693395 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1088.2839564285787 tokens/s 1088.0897563164995 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 785.0246977316517 tokens/s 784.884613075084 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.749286436480264 prompts/s 4.748476807201662 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 617.4072367424343 tokens/s 617.3019849362161 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 48.73236875033617 prompts/s 48.79581921111015 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3167.603968771851 tokens/s 3171.72824872216 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.1999971003978 prompts/s 24.22708159873138 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3121.7996259513156 tokens/s 3125.293526236348 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - 2:4 Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.586756093508087 prompts/s 5.578648697507994 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - 2:4 Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2612.1436790806415 tokens/s 2608.3529850068376 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.041494411059569 prompts/s 4.044106530230551 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4142.531771336058 tokens/s 4145.209193486315 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 14.157172819494397 prompts/s 14.13566061315486 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3638.39341461006 tokens/s 3632.864777580799 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 9.398865128295048 prompts/s 9.392961544017185 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2415.5083379718276 tokens/s 2413.9911168124167 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.996980796452526 prompts/s 24.758579707575098 prompts/s 1.03
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1559.803751769414 tokens/s 1609.3076809923814 tokens/s 1.03
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.49191812313221667 prompts/s 0.4918972664410856 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 130.03363666877016 tokens/s 130.02812341103657 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.99569453678974 tokens/s 121.99052207738923 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.731083187296554 prompts/s 2.7244396666690345 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 355.04081434855203 tokens/s 354.1771566669745 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.370576600941109 prompts/s 3.3672515848879416 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.1749581223442 tokens/s 437.7427060354324 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.783891008715171 prompts/s 5.780744121662275 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2967.1360874708826 tokens/s 2965.5217344127473 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.450388208688277 prompts/s 2.4510298349474677 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 756.745222528505 tokens/s 756.9433738273766 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 562.3412236040118 tokens/s 562.7597181437185 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.8172207793822974 prompts/s 1.8186241599053863 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 577.386769713657 tokens/s 577.832666742845 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 406.90117359460777 tokens/s 407.20328597998866 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.36654376222696 prompts/s 3.36660492732784 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 437.6506890895048 tokens/s 437.6586405526192 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.45020121648779 prompts/s 11.384394975067053 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1488.5261581434127 tokens/s 1479.971346758717 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.31260241353938 prompts/s 25.31894400415064 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1645.3191568800598 tokens/s 1645.7313602697916 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 15.865984974990951 prompts/s 15.85819651376316 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4077.5581385726746 tokens/s 4075.5565040371316 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.98358566254762 prompts/s 0.983631103681614 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.0135112210409 tokens/s 290.0269096908695 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.2971294042783 tokens/s 212.30693741863956 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.040590441499354 prompts/s 2.0384319394042425 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4181.169814632176 tokens/s 4176.747043839293 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.156934225747804 prompts/s 10.152973049137339 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1320.4014493472146 tokens/s 1319.886496387854 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.9284432526478152 prompts/s 2.926136100906273 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 380.69762284421597 tokens/s 380.3976931178155 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.070601799895014 prompts/s 7.079761182632859 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3305.930577558913 tokens/s 3310.213138551819 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.350491940782556 prompts/s 6.344786048806903 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 825.5639523017322 tokens/s 824.8221863448973 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.839209729532364 prompts/s 25.81640908488198 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1679.5486324196038 tokens/s 1678.0665905173287 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.3739693534489783 prompts/s 3.372912386549949 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.6160159483672 tokens/s 438.47861025149336 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.542726366362004 prompts/s 11.50371467682034 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1500.5544276270605 tokens/s 1495.4829079866442 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.156099211555572 prompts/s 10.14320166242732 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1320.2928975022246 tokens/s 1318.6162161155517 tokens/s 1.00

This comment was automatically generated by workflow using github-action-benchmark.

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

smaller_is_better

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 14671.440407500086 ms 14700.193791500169 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 208.6643606973424 ms 208.177746926672 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 161.46753849898232 ms 164.7100544996647 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 117.89186615758388 ms 118.11084162295658 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 124.39165958450867 ms 125.06230301204924 ms 0.99
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6387.269492000087 ms 6367.810279999503 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 111.88932599993375 ms 111.16892662006042 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 72.5244825007394 ms 69.513698499577 ms 1.04
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 36.289724544148655 ms 36.28406362460848 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.01978473213134 ms 36.9687584896116 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1923.126683501323 ms 1925.8178315003533 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 79.84467063327126 ms 79.36294155328748 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.41554399918823 ms 37.76919200026896 ms 1.07
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.005281552743925 ms 10.98752899201461 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.09402611059665 ms 11.090169800305919 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2136.2335784997413 ms 2124.492699500479 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 136.25811479666177 ms 136.58727046664958 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 104.31731850076176 ms 102.26584150041163 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 15.10960127592157 ms 15.11269000745712 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.650743723656545 ms 13.671886747409213 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 18574.00114200027 ms 18630.98225700014 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1102.113974073327 ms 1066.207715544699 ms 1.03
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 230.7276385017758 ms 232.56485849924502 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 162.38298922781706 ms 162.23560618336472 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 148.92584730583258 ms 149.11634172076177 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6134.139074499217 ms 6157.483541000147 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 106.36771252658946 ms 105.50076459997702 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 68.1048379992717 ms 63.466036500358314 ms 1.07
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 34.66033560242653 ms 34.73502434048436 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 35.17876881436496 ms 35.23139902220215 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 65613.32472899994 ms 66556.05931600257 ms 0.99
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 28628.61416851538 ms 29395.456832029286 ms 0.97
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 27332.809500499934 ms 28430.397764501322 ms 0.96
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 188.24308156207644 ms 188.6404633420125 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 199.30614640370118 ms 200.07395705000116 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6062.0934375001525 ms 6064.388077998956 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.8672970640667 ms 122.99509105597099 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 94.20262949970493 ms 91.56941200126312 ms 1.03
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 55.13576839101936 ms 55.034764851352534 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 48.08502645026652 ms 48.14404020383918 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 77644.25305650092 ms 78389.37180349967 ms 0.99
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 58056.55871913598 ms 58591.156928930686 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 71003.32071900084 ms 72002.09738900048 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 68.41126599025908 ms 68.32876132892734 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 65.81709570907654 ms 65.84333552122904 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 66410.8578134992 ms 66787.84569050003 ms 0.99
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23003.429712153353 ms 23200.36978493534 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 17349.811404999855 ms 17407.502998499695 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 261.2103598160443 ms 261.4966492002461 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 278.55816558679993 ms 279.02716963159884 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5222.535258501011 ms 5235.645166498216 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 89.90427876022295 ms 91.47958961346981 ms 0.98
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 61.31367800117005 ms 61.74462950002635 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.250890948022324 ms 37.26582722049259 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.679306717872336 ms 31.708689124541355 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7859.990413498963 ms 7866.947419000098 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 156.30002720930983 ms 156.33848048406313 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 127.74674149932252 ms 125.77558449993376 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 58.084073885518116 ms 57.926682457080105 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 57.587868500857205 ms 57.515965234391565 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5386.467451000499 ms 5385.054345500976 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 182.94118360667682 ms 183.15599952013872 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 172.65122749995498 ms 174.65145149981254 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.44215306390788 ms 40.29276032413655 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.65400143078365 ms 37.395030997344755 ms 1.01
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12297.73155500061 ms 12239.373073500246 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 198.2033804333405 ms 196.4258269800266 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 171.63947199969698 ms 172.22716749984102 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 96.09748940261052 ms 95.72504978387069 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 101.07422222877577 ms 100.80430680167527 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6068.778347000261 ms 6079.410813499635 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.58755756997077 ms 122.25182649330236 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 81.96738949936844 ms 83.33002849940385 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.834716322933176 ms 39.852174955426804 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.42931719131392 ms 39.52926163180674 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6054.284574499434 ms 6061.940364000293 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 127.06042101664936 ms 128.1418428899633 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 92.53861100114591 ms 87.2189094998248 ms 1.06
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.60141152783043 ms 40.458295919317536 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.35647327143947 ms 40.42338318231981 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1814.8084434997145 ms 1813.6266555002294 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 93.60767734337666 ms 93.09651339993555 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 57.111170000098355 ms 60.07437150037731 ms 0.95
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12.478057171616896 ms 12.492343204249421 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.78352964388927 ms 11.77166285193115 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 62489.75667349987 ms 64089.48246949967 ms 0.98
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40302.325046974685 ms 41101.79994840606 ms 0.98
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39308.59702949965 ms 40744.35401400115 ms 0.96
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.83318207948984 ms 99.1521024514054 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 102.22480738465657 ms 100.95686430257042 ms 1.01
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2070.6088160004583 ms 2067.7271800004746 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 116.26469750008255 ms 116.29355367333726 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 86.99318649996712 ms 86.71847000005073 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.632297370645555 ms 11.620019052349596 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.61245543034712 ms 11.619994200319992 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2426.99449849988 ms 2411.166190499898 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 116.47309425734177 ms 116.23203609204211 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 80.91874300043855 ms 80.91534550021606 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 18.093533888949484 ms 18.117534204784953 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 16.203817817727327 ms 16.20924310544079 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 258293.34289049986 ms 257960.56549199967 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 243556.67614118863 ms 243138.74433172468 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 244923.2257779995 ms 244561.7894119996 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 69.73269317950292 ms 70.2651706463995 ms 0.99
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 66.83747040314358 ms 66.51669527142957 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3653.525684001579 ms 3651.734495499113 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 143.22962458001712 ms 143.39316453994493 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 92.1650234995468 ms 94.5331505008653 ms 0.97
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.40717868159896 ms 24.4685511837163 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.733869918782684 ms 23.80589036230062 ms 1.00

This comment was automatically generated by workflow using github-action-benchmark.

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

bigger_is_better

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.359651000858093 prompts/s 3.3618719190890656 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1064.639805661921 tokens/s 1065.343592440134 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 767.9669439148141 tokens/s 768.4678895051088 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.693282240964379 prompts/s 3.6969976529082196 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 480.1266913253693 tokens/s 480.6096948780686 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.759064151534348 prompts/s 3.7554947864560835 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3853.0407553227064 tokens/s 3849.3821561174855 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.371490848281359 prompts/s 3.3672515848879416 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.29381027657666 tokens/s 437.7427060354324 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9680451591888342 prompts/s 0.9680309403748015 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 288.64847874972924 tokens/s 288.64423903115704 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 210.0464386407932 tokens/s 210.04012667272318 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9836178627518221 prompts/s 0.983631103681614 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.0230055585839 tokens/s 290.0269096908695 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.30407949635327 tokens/s 212.30693741863956 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 16.808540809974144 prompts/s 16.74529333683377 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2185.1103052966387 tokens/s 2176.8881337883904 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.46397632040384684 prompts/s 0.4639583106048493 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.64750053555287 tokens/s 122.64273982528587 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 115.07850016203145 tokens/s 115.07403325161876 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 48.62428052191339 prompts/s 48.79581921111015 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3160.57823392437 tokens/s 3171.72824872216 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - 2:4 Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.5923154256092875 prompts/s 5.578648697507994 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - 2:4 Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2614.7430003978784 tokens/s 2608.3529850068376 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.568281724227308 prompts/s 5.581164920707232 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 723.87662414955 tokens/s 725.5514396919401 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9474643960583496 prompts/s 0.9475222154869255 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 279.3630353924579 tokens/s 279.38008364370495 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 204.5259809624624 tokens/s 204.5353038423927 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.069274098818043 prompts/s 2.0683856916680297 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4239.94262847817 tokens/s 4238.122282227793 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0024406146660225 prompts/s 2.002675145423224 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4103.00081945068 tokens/s 4103.481372972186 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.44444249539314 prompts/s 3.43365128693395 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1091.509382365132 tokens/s 1088.0897563164995 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 787.3490359569394 tokens/s 784.884613075084 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4658861140142124 prompts/s 0.465998128153441 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.98181704429656 tokens/s 126.01210716106583 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.05243538377466 tokens/s 118.08081901989425 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7950747547125352 prompts/s 1.7943553898000457 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.3597181126296 tokens/s 233.26620067400594 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 8.021179764753187 prompts/s 8.014925201315489 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4114.865219318385 tokens/s 4111.656628274846 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.467169124441677 prompts/s 3.47554536374714 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1098.711223844323 tokens/s 1101.365570317831 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 792.62028775428 tokens/s 794.5305234247786 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.524531039249862 prompts/s 11.50371467682034 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1498.1890351024822 tokens/s 1495.4829079866442 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.75848529171824 prompts/s 4.748476807201662 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 618.6030879233713 tokens/s 617.3019849362161 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.8118496087665914 prompts/s 1.8186241599053863 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 575.6801840931482 tokens/s 577.832666742845 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 405.68279060075326 tokens/s 407.20328597998866 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.489309820721081 prompts/s 3.4889400599782956 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3576.542566239108 tokens/s 3576.163561477753 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.587029066746503 prompts/s 5.583551563685145 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2612.271310447995 tokens/s 2610.645369116626 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.647967207720695 prompts/s 6.633908223317837 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 864.2357370036905 tokens/s 862.4080690313189 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.75890153029498 prompts/s 24.758579707575098 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1609.328599469174 tokens/s 1609.3076809923814 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.962038763211396 prompts/s 23.909559126416504 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3091.10300045427 tokens/s 3084.3331273077292 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.046575337515022 prompts/s 4.044106530230551 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4147.739720952898 tokens/s 4145.209193486315 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 19.702793355245678 prompts/s 19.663083685272863 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2561.363136181938 tokens/s 2556.2008790854725 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.419570074809217 prompts/s 11.384394975067053 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1484.544109725198 tokens/s 1479.971346758717 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.268905687991 prompts/s 23.549919588363764 prompts/s 1.01
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3001.688833750839 tokens/s 3037.9396268989253 tokens/s 1.01
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.467160622349648 prompts/s 7.459659949219531 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3830.6533992653694 tokens/s 3826.8055539496195 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 28.735096292116673 prompts/s 28.794803850253647 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3706.827421683051 tokens/s 3714.5296966827204 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.9741659172373591 prompts/s 1.975317062191078 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 256.6415692408567 tokens/s 256.7912180848401 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.177143609880815 prompts/s 24.194691196838754 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3118.851525674625 tokens/s 3121.115164392199 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.847544352383526 prompts/s 13.888661076110504 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3558.818898562566 tokens/s 3569.3858965603995 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 14.167830848995997 prompts/s 14.13566061315486 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3641.1325281919712 tokens/s 3632.864777580799 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.29145245042784 prompts/s 2.290440902922561 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 707.661622090795 tokens/s 707.3492292465648 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 458.7610016553892 tokens/s 458.4974060258344 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.774533096750673 prompts/s 25.31894400415064 prompts/s 0.98
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1675.3446512887938 tokens/s 1645.7313602697916 tokens/s 0.98
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.24043759193560643 prompts/s 0.24029023094156635 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.256886951628836 tokens/s 31.237730022403625 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9133276283846216 prompts/s 0.9127050864021167 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.73259169000082 tokens/s 118.65166123227517 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.351624843791534 prompts/s 6.344786048806903 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 825.7112296928995 tokens/s 824.8221863448973 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.045998686964363 prompts/s 6.036791955001028 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2826.8671460770574 tokens/s 2822.5624464802804 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9839553340237884 prompts/s 0.9839491630042055 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.12251008802747 tokens/s 290.12069054233336 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.40971780682864 tokens/s 212.39854615663114 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.394848322054619 prompts/s 11.415719977955193 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1481.3302818671004 tokens/s 1484.0435971341751 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0396869809873426 prompts/s 2.0384319394042425 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4179.318624043065 tokens/s 4176.747043839293 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.491888759465498 prompts/s 0.4918873995766033 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 130.02587467710973 tokens/s 130.0255152040793 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.46390481492682 tokens/s 121.99791284298915 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.08471199179297 prompts/s 7.079761182632859 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3312.5279388827207 tokens/s 3310.213138551819 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.8658447195294516 prompts/s 3.8650348835012154 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3962.4908375176883 tokens/s 3961.660755588746 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9125956792858986 prompts/s 0.9123163639924358 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.63743830716682 tokens/s 118.60112731901664 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.593488374213245 prompts/s 5.583649217567208 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2615.291424247145 tokens/s 2610.691028165724 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.181613457043313 prompts/s 3.179495041356072 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1499.7044087926781 tokens/s 1498.7058596638462 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.369400752362948 prompts/s 3.36660492732784 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.0220978071833 tokens/s 437.6586405526192 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.141356803997157 prompts/s 10.152973049137339 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1318.3763845196304 tokens/s 1319.886496387854 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0386685476113153 prompts/s 2.036458994186228 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4177.231854055585 tokens/s 4172.704479087582 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.3876774458937 prompts/s 4.390290227894882 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1390.4111058292547 tokens/s 1391.239070317609 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1002.8738830981395 tokens/s 1003.5588820335794 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12.2376184242936 prompts/s 12.264694778738873 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1578.6527767338744 tokens/s 1582.1456264573146 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.462054297330851 prompts/s 7.4600015789981065 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3828.0338545307263 tokens/s 3826.980810026029 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.760200238822491 prompts/s 3.752344427261337 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3854.205244793053 tokens/s 3846.1530379428705 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.47410938572688893 prompts/s 0.47413571902199947 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.32607502304582 tokens/s 125.33303596627533 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.24689807882875 tokens/s 99.25241051527189 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.9350355022811305 prompts/s 2.926136100906273 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 381.55461529654696 tokens/s 380.3976931178155 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.7298696460644165 prompts/s 2.7244396666690345 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 354.8830539883742 tokens/s 354.1771566669745 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.3701862916468053 prompts/s 3.372912386549949 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.1242179140847 tokens/s 438.47861025149336 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.155451662985653 prompts/s 10.14320166242732 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1320.208716188135 tokens/s 1318.6162161155517 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 15.856531121109779 prompts/s 15.85819651376316 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4075.128498125213 tokens/s 4075.5565040371316 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.038907542900164 prompts/s 2.037665339068763 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4177.721555402436 tokens/s 4175.176279751895 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.7652391915447462 prompts/s 0.7639268340372671 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.48109490081701 tokens/s 99.31048842484472 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9122808922611173 prompts/s 0.9127306469948456 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.59651599394525 tokens/s 118.65498410932993 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.9379461353071608 prompts/s 1.938586235550944 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3970.851631244372 tokens/s 3972.163196643884 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 8.197421317534967 prompts/s 8.19876334119261 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4205.277135895439 tokens/s 4205.965594031809 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.403701275303967 prompts/s 6.3452258756993265 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 832.4811657895157 tokens/s 824.8793638409124 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.636225036009988 prompts/s 6.631018479858359 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 862.7092546812985 tokens/s 862.0324023815867 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9839161960039398 prompts/s 0.9838806054260277 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.110970112415 tokens/s 290.10047611188236 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.7161221347451 tokens/s 212.40342470138896 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7939318679498872 prompts/s 1.7931727190644227 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.21114283348535 tokens/s 233.11245347837496 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.2405907573046605 prompts/s 0.24030861859274258 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.276798449605867 tokens/s 31.240120417056538 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.609371750142447 prompts/s 25.284467324256294 prompts/s 1.03
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1599.609163759259 tokens/s 1643.490376076659 tokens/s 1.03
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.450669876450265 prompts/s 2.4510298349474677 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 756.8322090445471 tokens/s 756.9433738273766 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 562.84698430425 tokens/s 562.7597181437185 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - synthetic\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.8035710604267052 prompts/s 3.803884961395343 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - synthetic\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1460.5712872038548 tokens/s 1460.6918251758116 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 9.393011833959958 prompts/s 9.392961544017185 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2414.0040413277093 tokens/s 2413.9911168124167 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.277767581244198 prompts/s 6.262518754196035 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 816.1097855617458 tokens/s 814.1274380454846 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.2449122117895 prompts/s 24.22708159873138 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3127.5936753208453 tokens/s 3125.293526236348 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.0138660495526803 prompts/s 1.0135982483583388 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 131.80258644184846 tokens/s 131.76777228658406 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4919047333430463 prompts/s 0.4918972664410856 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 130.03009721190085 tokens/s 130.02812341103657 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.99237386907548 tokens/s 121.99052207738923 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.119093124027047 prompts/s 4.119903332561249 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1305.2994200729308 tokens/s 1305.5561670553343 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 821.9952395825067 tokens/s 822.7144828880427 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.23756520354147248 prompts/s 0.2402768434229402 prompts/s 1.01
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 30.88347646039142 tokens/s 31.235989644982226 tokens/s 1.01
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.0781451737460985 prompts/s 7.051461020667885 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 920.1588725869929 tokens/s 916.689932686825 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.2252024719972945 prompts/s 2.2244860054437767 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 687.2018620853512 tokens/s 686.9805981078501 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 510.76110767574164 tokens/s 510.5996197455427 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.824482100442037 prompts/s 25.81640908488198 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1678.5913365287324 tokens/s 1678.0665905173287 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.280369801855608 prompts/s 13.304553986613938 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 863.2240371206145 tokens/s 864.796009129906 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.648951254966757 prompts/s 6.652228260231785 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 864.3636631456784 tokens/s 864.7896738301321 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.472777940328911 prompts/s 7.460202709235207 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3833.5350833887314 tokens/s 3827.083989837661 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.761220779420484 prompts/s 3.7569492477931656 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3855.251298905996 tokens/s 3850.8729789879944 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.680711880746224 prompts/s 1.6806087990413068 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 548.1428908882249 tokens/s 548.1092720958677 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 382.0078829002221 tokens/s 381.9844535282326 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7939315602097863 prompts/s 1.7930718518766309 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.2111028272722 tokens/s 233.09934074396202 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.3517792760199225 prompts/s 2.35158613748706 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 726.2921545489793 tokens/s 726.2325082196705 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 539.7113939066627 tokens/s 539.6984248978303 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.7748042418941027 prompts/s 0.7735157651280619 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 100.72455144623335 tokens/s 100.55704946664804 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.256349363313981 prompts/s 0.2562593094265349 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 33.325417230817536 tokens/s 33.31371022544953 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.216736285181321 prompts/s 2.2167695683344157 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 684.5872778315968 tokens/s 684.5975565568232 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 508.81782430699946 tokens/s 508.8165968661773 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.78361867733124 prompts/s 5.780744121662275 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2966.996381470926 tokens/s 2965.5217344127473 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.861120444451192 prompts/s 13.862772772855124 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3562.307954223956 tokens/s 3562.732602623767 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.812264392948462 prompts/s 13.854177055679637 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3549.7519489877545 tokens/s 3560.523503309667 tokens/s 1.00

This comment was automatically generated by workflow using github-action-benchmark.

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

smaller_is_better

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 66931.1053829997 ms 66787.84569050003 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23342.27604626999 ms 23200.36978493534 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 17800.132582000515 ms 17407.502998499695 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 261.657417060096 ms 261.4966492002461 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 278.97989142772093 ms 279.02716963159884 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6064.163899500272 ms 6079.410813499635 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.84322424331792 ms 122.25182649330236 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 81.91542849999678 ms 83.33002849940385 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.81237977005924 ms 39.852174955426804 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.49480608319446 ms 39.52926163180674 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3666.834318499241 ms 3651.734495499113 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 143.11174139328313 ms 143.39316453994493 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 91.59698149960604 ms 94.5331505008653 ms 0.97
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.36734761260045 ms 24.4685511837163 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.649514210886167 ms 23.80589036230062 ms 0.99
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6355.360378500336 ms 6367.810279999503 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 111.07957002678936 ms 111.16892662006042 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 69.13049949980632 ms 69.513698499577 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 36.273124083950954 ms 36.28406362460848 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 36.95699933018845 ms 36.9687584896116 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6063.323772000331 ms 6061.940364000293 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 127.74053738003205 ms 128.1418428899633 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 92.22327899988159 ms 87.2189094998248 ms 1.06
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.47740103380456 ms 40.458295919317536 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.41175559645828 ms 40.42338318231981 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 62434.5530745004 ms 64089.48246949967 ms 0.97
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40341.59530111335 ms 41101.79994840606 ms 0.98
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40089.88794800007 ms 40744.35401400115 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.23333490961022 ms 99.1521024514054 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 101.38731865845863 ms 100.95686430257042 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6138.0041939992225 ms 6157.483541000147 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 105.25370153986538 ms 105.50076459997702 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 69.34573000035016 ms 63.466036500358314 ms 1.09
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 34.66916330861514 ms 34.73502434048436 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 35.20249453031695 ms 35.23139902220215 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 67696.56554000176 ms 66556.05931600257 ms 1.02
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 30013.547229894015 ms 29395.456832029286 ms 1.02
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 29051.331959499294 ms 28430.397764501322 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 189.95778387274478 ms 188.6404633420125 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 200.71795366500223 ms 200.07395705000116 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 257689.3706384999 ms 257960.56549199967 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 243934.78863560208 ms 243138.74433172468 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 244595.2843140003 ms 244561.7894119996 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 69.76695993422113 ms 70.2651706463995 ms 0.99
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 67.07681470882945 ms 66.51669527142957 ms 1.01
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6068.786497999099 ms 6064.388077998956 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.80484077736523 ms 122.99509105597099 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 89.96812399891496 ms 91.56941200126312 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 55.036998853002125 ms 55.034764851352534 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 48.161595403597964 ms 48.14404020383918 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2142.5964224999916 ms 2124.492699500479 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 136.95423646659037 ms 136.58727046664958 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 105.06599900054425 ms 102.26584150041163 ms 1.03
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 15.088356492075098 ms 15.11269000745712 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.655097184058752 ms 13.671886747409213 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1923.5307070002818 ms 1925.8178315003533 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 79.52589756001544 ms 79.36294155328748 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.84988749976037 ms 37.76919200026896 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.990196552600976 ms 10.98752899201461 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.101862315424263 ms 11.090169800305919 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 14727.956755001287 ms 14700.193791500169 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 207.43928320336272 ms 208.177746926672 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 161.8715450003947 ms 164.7100544996647 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.73434108304961 ms 118.11084162295658 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.11435313409407 ms 125.06230301204924 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5239.008678499886 ms 5235.645166498216 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 90.91267326681798 ms 91.47958961346981 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 61.70038650088827 ms 61.74462950002635 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.28207667565892 ms 37.26582722049259 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.687763794898334 ms 31.708689124541355 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1815.862186999766 ms 1813.6266555002294 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 92.85708308672838 ms 93.09651339993555 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 59.3996994994086 ms 60.07437150037731 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12.503594600157154 ms 12.492343204249421 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.726245442219327 ms 11.77166285193115 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2432.085828500931 ms 2411.166190499898 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 115.97731625733529 ms 116.23203609204211 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 83.34300799924677 ms 80.91534550021606 ms 1.03
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 18.038354873429853 ms 18.117534204784953 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 16.147303457152088 ms 16.20924310544079 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2071.159581500069 ms 2067.7271800004746 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 117.38934138680634 ms 116.29355367333726 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 86.24154350036406 ms 86.71847000005073 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.614447459022692 ms 11.620019052349596 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.609165830648937 ms 11.619994200319992 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 18468.7140394999 ms 18630.98225700014 ms 0.99
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1110.133476052722 ms 1066.207715544699 ms 1.04
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 236.71987249872473 ms 232.56485849924502 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 162.36971628466682 ms 162.23560618336472 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 149.287468383281 ms 149.11634172076177 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7876.454225500311 ms 7866.947419000098 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 156.95443960801884 ms 156.33848048406313 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 128.45143799859216 ms 125.77558449993376 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 57.94276547439565 ms 57.926682457080105 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 57.508369824266325 ms 57.515965234391565 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 78565.39773550048 ms 78389.37180349967 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 58711.33112383332 ms 58591.156928930686 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 72157.52103149952 ms 72002.09738900048 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 68.35338342569004 ms 68.32876132892734 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 65.91079877096972 ms 65.84333552122904 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5395.843203001277 ms 5385.054345500976 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 183.65219166668734 ms 183.15599952013872 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 178.78151750119287 ms 174.65145149981254 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.48714043924613 ms 40.29276032413655 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.688588111216575 ms 37.395030997344755 ms 1.01
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12238.293557499674 ms 12239.373073500246 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 199.42883660927086 ms 196.4258269800266 ms 1.02
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 168.41479350023292 ms 172.22716749984102 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 94.97464477132647 ms 95.72504978387069 ms 0.99
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.14474897894794 ms 100.80430680167527 ms 0.98

This comment was automatically generated by workflow using github-action-benchmark.

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

bigger_is_better

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.3511531038449704 prompts/s 2.35158613748706 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 726.0987758834294 tokens/s 726.2325082196705 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 539.5990419448361 tokens/s 539.6984248978303 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.472686136591147 prompts/s 7.460202709235207 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3833.4879880712583 tokens/s 3827.083989837661 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 8.025589624567663 prompts/s 8.014925201315489 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4117.127477403211 tokens/s 4111.656628274846 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.389714830256111 prompts/s 4.390290227894882 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1391.056732559859 tokens/s 1391.239070317609 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1002.4235727310714 tokens/s 1003.5588820335794 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0711685620946594 prompts/s 2.0683856916680297 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4243.824383731958 tokens/s 4238.122282227793 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.046420863823796 prompts/s 4.044106530230551 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4147.581385419391 tokens/s 4145.209193486315 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.7607293137087012 prompts/s 3.752344427261337 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3854.747546551419 tokens/s 3846.1530379428705 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0406487975953667 prompts/s 2.0384319394042425 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4181.289386272906 tokens/s 4176.747043839293 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.801426492924243 prompts/s 25.81640908488198 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1677.0927220400758 tokens/s 1678.0665905173287 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.240298761258473 prompts/s 0.24030861859274258 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.23883896360149 tokens/s 31.240120417056538 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.63899658064691 prompts/s 6.633908223317837 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 863.0695554840983 tokens/s 862.4080690313189 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.2163723973516714 prompts/s 2.2167695683344157 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 684.4748995661255 tokens/s 684.5975565568232 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 508.7254339457274 tokens/s 508.8165968661773 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0414191617628297 prompts/s 2.037665339068763 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4182.867862452038 tokens/s 4175.176279751895 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.17195958330743 prompts/s 10.152973049137339 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1322.3547458299659 tokens/s 1319.886496387854 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.7646771783950157 prompts/s 0.7639268340372671 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.40803319135203 tokens/s 99.31048842484472 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.223950435991538 prompts/s 2.2244860054437767 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 686.8151999791467 tokens/s 686.9805981078501 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 510.4826178102496 tokens/s 510.5996197455427 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.586875646395628 prompts/s 5.581164920707232 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 726.2938340314315 tokens/s 725.5514396919401 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.6386794862732845 prompts/s 6.631018479858359 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 863.0283332155269 tokens/s 862.0324023815867 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.3688326679552905 prompts/s 3.36660492732784 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 437.9482468341878 tokens/s 437.6586405526192 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 14.139313210774919 prompts/s 14.13566061315486 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3633.8034951691543 tokens/s 3632.864777580799 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.774849889081665 prompts/s 23.549919588363764 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3066.955635691535 tokens/s 3037.9396268989253 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 28.81886072317098 prompts/s 28.794803850253647 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3717.6330332890566 tokens/s 3714.5296966827204 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - synthetic\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.8009732297188745 prompts/s 3.803884961395343 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - synthetic\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1459.5737202120479 tokens/s 1460.6918251758116 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.793187728191505 prompts/s 1.7931727190644227 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.11440466489563 tokens/s 233.11245347837496 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.23615010499036587 prompts/s 0.2402768434229402 prompts/s 1.02
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 30.699513648747562 tokens/s 31.235989644982226 tokens/s 1.02
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.7737684204481046 prompts/s 0.7735157651280619 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 100.5898946582536 tokens/s 100.55704946664804 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.7304495252436247 prompts/s 2.7244396666690345 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 354.95843828167125 tokens/s 354.1771566669745 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4658672395317282 prompts/s 0.465998128153441 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.97671313257307 tokens/s 126.01210716106583 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.04765271574306 tokens/s 118.08081901989425 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.46542175277292 prompts/s 7.459659949219531 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3829.7613591725076 tokens/s 3826.8055539496195 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.041861241761859 prompts/s 6.036791955001028 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2824.932642198175 tokens/s 2822.5624464802804 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.4503314314703633 prompts/s 2.4510298349474677 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 756.7276882095541 tokens/s 756.9433738273766 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 562.5895624484448 tokens/s 562.7597181437185 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.346473920194909 prompts/s 6.344786048806903 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 825.0416096253382 tokens/s 824.8221863448973 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - 2:4 Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.591688507590615 prompts/s 5.578648697507994 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - 2:4 Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2614.449878609068 tokens/s 2608.3529850068376 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.138587913022405 prompts/s 10.14320166242732 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1318.0164286929125 tokens/s 1318.6162161155517 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.348431409794471 prompts/s 6.3452258756993265 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 825.2960832732813 tokens/s 824.8793638409124 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.6845929782495725 prompts/s 1.6806087990413068 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 549.4086616783736 tokens/s 548.1092720958677 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 382.90349170818587 tokens/s 381.9844535282326 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0411120336752067 prompts/s 2.036458994186228 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4182.238557000499 tokens/s 4172.704479087582 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 9.39092630642859 prompts/s 9.392961544017185 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2413.4680607521473 tokens/s 2413.9911168124167 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 48.91241532800159 prompts/s 48.79581921111015 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3179.3069963201033 tokens/s 3171.72824872216 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.834401330884333 prompts/s 13.888661076110504 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3555.4411420372735 tokens/s 3569.3858965603995 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7937210557221104 prompts/s 1.7943553898000457 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.18373724387436 tokens/s 233.26620067400594 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.9978414600364625 prompts/s 7.051461020667885 prompts/s 1.01
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 909.7193898047401 tokens/s 916.689932686825 tokens/s 1.01
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.79411542012935 prompts/s 24.22708159873138 prompts/s 1.02
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3069.4408891966864 tokens/s 3125.293526236348 tokens/s 1.02
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.852344672102532 prompts/s 13.854177055679637 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3560.052580730351 tokens/s 3560.523503309667 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.586979006933388 prompts/s 5.583649217567208 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2612.2479044817746 tokens/s 2610.691028165724 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.520862567206194 prompts/s 11.50371467682034 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1497.7121337368053 tokens/s 1495.4829079866442 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4741050230865814 prompts/s 0.47413571902199947 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.32492180270692 tokens/s 125.33303596627533 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.24598483279104 tokens/s 99.25241051527189 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.1812403207567 prompts/s 3.179495041356072 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1499.5285250338027 tokens/s 1498.7058596638462 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.49190726824952813 prompts/s 0.4918972664410856 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 130.03076728908027 tokens/s 130.02812341103657 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.99300252588299 tokens/s 121.99052207738923 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 16.72670253946786 prompts/s 16.74529333683377 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2174.4713301308216 tokens/s 2176.8881337883904 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.2564868306193423 prompts/s 0.2562593094265349 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 33.343287980514496 tokens/s 33.31371022544953 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.271043805389769 prompts/s 6.262518754196035 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 815.2356947006699 tokens/s 814.1274380454846 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.354780093746995 prompts/s 3.3618719190890656 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1063.0962639074853 tokens/s 1065.343592440134 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 766.8423433888755 tokens/s 768.4678895051088 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.118958433135383 prompts/s 4.119903332561249 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1305.2567378762712 tokens/s 1305.5561670553343 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 821.7349533827976 tokens/s 822.7144828880427 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.78908341349774 prompts/s 25.31894400415064 prompts/s 0.98
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1676.290421877353 tokens/s 1645.7313602697916 tokens/s 0.98
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.4548671489274376 prompts/s 3.43365128693395 prompts/s 0.99
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1094.8128508236157 tokens/s 1088.0897563164995 tokens/s 0.99
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 789.7296556151953 tokens/s 784.884613075084 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.912167866077675 prompts/s 0.9127306469948456 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.58182259009776 tokens/s 118.65498410932993 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9679789020617173 prompts/s 0.9680309403748015 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 288.62872242042266 tokens/s 288.64423903115704 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 210.02883557301124 tokens/s 210.04012667272318 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.762205327749965 prompts/s 3.7569492477931656 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3856.2604609437144 tokens/s 3850.8729789879944 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.779383400969849 prompts/s 5.780744121662275 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2964.8236846975324 tokens/s 2965.5217344127473 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.983973383312484 prompts/s 0.9838806054260277 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.1278319809636 tokens/s 290.10047611188236 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.6071289323284 tokens/s 212.40342470138896 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.9752272494452592 prompts/s 1.975317062191078 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 256.7795424278837 tokens/s 256.7912180848401 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.648436725188829 prompts/s 6.652228260231785 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 864.2967742745477 tokens/s 864.7896738301321 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.74291500259038 prompts/s 4.748476807201662 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 616.5789503367494 tokens/s 617.3019849362161 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.411910971880014 prompts/s 11.384394975067053 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1483.5484263444018 tokens/s 1479.971346758717 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.2402490152599145 prompts/s 0.24029023094156635 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.232371983788884 tokens/s 31.237730022403625 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.588085026975937 prompts/s 5.583551563685145 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2612.7650352128685 tokens/s 2610.645369116626 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.9385333640406757 prompts/s 1.938586235550944 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3972.0548629193445 tokens/s 3972.163196643884 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4918939582156238 prompts/s 0.4918873995766033 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 130.027248914718 tokens/s 130.0255152040793 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.99298093052947 tokens/s 121.99791284298915 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.7636346084778496 prompts/s 3.7554947864560835 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3857.725473689796 tokens/s 3849.3821561174855 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.3730185038380274 prompts/s 3.372912386549949 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.4924054989435 tokens/s 438.47861025149336 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.46381598415931496 prompts/s 0.4639583106048493 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.60511725267332 tokens/s 122.64273982528587 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 115.03873249775437 tokens/s 115.07403325161876 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.8145070645327142 prompts/s 1.8186241599053863 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 576.5245392853557 tokens/s 577.832666742845 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 406.27780911988555 tokens/s 407.20328597998866 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9836357287490266 prompts/s 0.983631103681614 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.028273407413 tokens/s 290.0269096908695 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.3079356931899 tokens/s 212.30693741863956 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.4914362139823942 prompts/s 3.4889400599782956 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3578.722119331954 tokens/s 3576.163561477753 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 8.204547777579114 prompts/s 8.19876334119261 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4208.933009898085 tokens/s 4205.965594031809 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.076754423827196 prompts/s 7.079761182632859 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3308.8072984046435 tokens/s 3310.213138551819 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.911390185924294 prompts/s 25.284467324256294 prompts/s 1.01
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1619.2403620850791 tokens/s 1643.490376076659 tokens/s 1.01
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.8647836193787968 prompts/s 3.8650348835012154 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3961.4032098632665 tokens/s 3961.660755588746 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12.23935707742525 prompts/s 12.264694778738873 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1578.8770629878572 tokens/s 1582.1456264573146 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.22192339324607 prompts/s 24.194691196838754 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3124.628117728743 tokens/s 3121.115164392199 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9474371837606016 prompts/s 0.9475222154869255 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 279.3550117557592 tokens/s 279.38008364370495 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 204.52326485840106 tokens/s 204.5353038423927 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7922852808298626 prompts/s 1.7930718518766309 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 232.99708650788213 tokens/s 233.09934074396202 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.40309009362813 prompts/s 11.415719977955193 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1482.401712171657 tokens/s 1484.0435971341751 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.0145673029253242 prompts/s 1.0135982483583388 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 131.89374938029212 tokens/s 131.76777228658406 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.291946871435903 prompts/s 2.290440902922561 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 707.8143124826452 tokens/s 707.3492292465648 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 458.6216249034861 tokens/s 458.4974060258344 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9127678014601871 prompts/s 0.9123163639924358 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.65981418982433 tokens/s 118.60112731901664 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9119941321421412 prompts/s 0.9127050864021167 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.55923717847836 tokens/s 118.65166123227517 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.890353978773994 prompts/s 23.909559126416504 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3081.855663261845 tokens/s 3084.3331273077292 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.28682819423898 prompts/s 13.304553986613938 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 863.6438326255337 tokens/s 864.796009129906 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0025984957188454 prompts/s 2.002675145423224 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4103.324317727914 tokens/s 4103.481372972186 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.54165754916716 prompts/s 24.758579707575098 prompts/s 1.01
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1595.2077406958651 tokens/s 1609.3076809923814 tokens/s 1.01
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 19.696341624271337 prompts/s 19.663083685272863 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2560.524411155274 tokens/s 2556.2008790854725 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 15.874550917132854 prompts/s 15.85819651376316 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4079.7595857031433 tokens/s 4075.5565040371316 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.471694754683738 prompts/s 7.4600015789981065 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3832.9794091527574 tokens/s 3826.980810026029 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.858851752797936 prompts/s 13.862772772855124 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3561.72490046907 tokens/s 3562.732602623767 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.721109273148228 prompts/s 2.926136100906273 prompts/s 1.08
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 353.74420550926965 tokens/s 380.3976931178155 tokens/s 1.08
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.4803217697460043 prompts/s 3.47554536374714 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1102.8791656148112 tokens/s 1101.365570317831 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 795.891583378082 tokens/s 794.5305234247786 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.6940803695188453 prompts/s 3.6969976529082196 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 480.2304480374499 tokens/s 480.6096948780686 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.370181587246997 prompts/s 3.3672515848879416 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.1236063421096 tokens/s 437.7427060354324 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9838849543798242 prompts/s 0.9839491630042055 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.10175841540575 tokens/s 290.12069054233336 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.39124510197266 tokens/s 212.39854615663114 tokens/s 1.00

This comment was automatically generated by workflow using github-action-benchmark.

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

smaller_is_better

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5403.3334414980345 ms 5385.054345500976 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 182.6672527786577 ms 183.15599952013872 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 173.26183499972103 ms 174.65145149981254 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.570686681999014 ms 40.29276032413655 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.71272982877713 ms 37.395030997344755 ms 1.01
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 14712.270560000434 ms 14700.193791500169 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 211.4012420333347 ms 208.177746926672 ms 1.02
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 165.1276724996933 ms 164.7100544996647 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.61316128523288 ms 118.11084162295658 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.23053596180378 ms 125.06230301204924 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12388.316153499545 ms 12239.373073500246 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 202.90839387460193 ms 196.4258269800266 ms 1.03
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 177.19953449977766 ms 172.22716749984102 ms 1.03
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 95.76608171097477 ms 95.72504978387069 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 100.6277065136866 ms 100.80430680167527 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7894.943926500673 ms 7866.947419000098 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 157.38730911329912 ms 156.33848048406313 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 131.29129100070713 ms 125.77558449993376 ms 1.04
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 58.145339403127906 ms 57.926682457080105 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 57.77174892640247 ms 57.515965234391565 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6143.9237764998325 ms 6157.483541000147 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 105.57952202662516 ms 105.50076459997702 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 68.90616300006513 ms 63.466036500358314 ms 1.09
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 34.740065797105615 ms 34.73502434048436 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 35.27362721178171 ms 35.23139902220215 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2431.906657499894 ms 2411.166190499898 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 116.1085367613535 ms 116.23203609204211 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 80.23967450026248 ms 80.91534550021606 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 18.105119612849144 ms 18.117534204784953 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 16.194528842399823 ms 16.20924310544079 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 78082.65650449993 ms 78389.37180349967 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 58163.472767429324 ms 58591.156928930686 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 71294.46411749996 ms 72002.09738900048 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 67.35790102715542 ms 68.32876132892734 ms 0.99
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 65.52902275067585 ms 65.84333552122904 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5234.401947000151 ms 5235.645166498216 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 90.68538420006614 ms 91.47958961346981 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 60.826272501799394 ms 61.74462950002635 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.25110775777427 ms 37.26582722049259 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.682919228131336 ms 31.708689124541355 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2065.334467500179 ms 2067.7271800004746 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 117.37230603336987 ms 116.29355367333726 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 86.2206050005625 ms 86.71847000005073 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.59097868551506 ms 11.620019052349596 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.618711720487603 ms 11.619994200319992 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 67333.70489449863 ms 66787.84569050003 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23449.42444652001 ms 23200.36978493534 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 17718.90502399947 ms 17407.502998499695 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 262.3360423793547 ms 261.4966492002461 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 279.50231642284837 ms 279.02716963159884 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 18693.89975500053 ms 18630.98225700014 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1106.037561508017 ms 1066.207715544699 ms 1.04
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 234.87662999832537 ms 232.56485849924502 ms 1.01
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 162.23267729135304 ms 162.23560618336472 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 149.22055482149233 ms 149.11634172076177 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 61705.24036599909 ms 64089.48246949967 ms 0.96
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39382.883654935344 ms 41101.79994840606 ms 0.96
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 38853.11500399985 ms 40744.35401400115 ms 0.95
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 98.6048637101342 ms 99.1521024514054 ms 0.99
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 100.93879064096944 ms 100.95686430257042 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6106.1727380001685 ms 6079.410813499635 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.2248624533313 ms 122.25182649330236 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 82.13618149875401 ms 83.33002849940385 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.866295219973715 ms 39.852174955426804 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.483253903939236 ms 39.52926163180674 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1841.4298955003687 ms 1813.6266555002294 ms 1.02
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 92.76224013333679 ms 93.09651339993555 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 59.39676350044465 ms 60.07437150037731 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12.53072772650871 ms 12.492343204249421 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.7504803795625 ms 11.77166285193115 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1928.7328335003622 ms 1925.8178315003533 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 79.24429737999162 ms 79.36294155328748 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 38.83016149939067 ms 37.76919200026896 ms 1.03
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.01682881146633 ms 10.98752899201461 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.104867589473509 ms 11.090169800305919 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6375.923833499655 ms 6367.810279999503 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 109.70417828670179 ms 111.16892662006042 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 70.59495649991732 ms 69.513698499577 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 36.29284477344846 ms 36.28406362460848 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 36.98958335851306 ms 36.9687584896116 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 260014.52012900062 ms 257960.56549199967 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 244655.706315512 ms 243138.74433172468 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 246336.86162399955 ms 244561.7894119996 ms 1.01
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 70.77495113818273 ms 70.2651706463995 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 66.8261656403469 ms 66.51669527142957 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3645.2240139988135 ms 3651.734495499113 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 143.44058583311076 ms 143.39316453994493 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 94.05171199978213 ms 94.5331505008653 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.43651480224753 ms 24.4685511837163 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.768530231824148 ms 23.80589036230062 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6056.600670500302 ms 6061.940364000293 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 128.3224846033469 ms 128.1418428899633 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 87.35010950022115 ms 87.2189094998248 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.54763684967656 ms 40.458295919317536 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.42104585396235 ms 40.42338318231981 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6055.8608940009435 ms 6064.388077998956 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.8573354986826 ms 122.99509105597099 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 92.14041750055912 ms 91.56941200126312 ms 1.01
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 55.07886411780935 ms 55.034764851352534 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 48.0826652607532 ms 48.14404020383918 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 66309.31746399801 ms 66556.05931600257 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 29019.913823834024 ms 29395.456832029286 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 27652.122268498715 ms 28430.397764501322 ms 0.97
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 188.22594207093397 ms 188.6404633420125 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 199.4164384190758 ms 200.07395705000116 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2122.4544235001304 ms 2124.492699500479 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 136.2453950600563 ms 136.58727046664958 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 104.65170849965943 ms 102.26584150041163 ms 1.02
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 15.058391072166453 ms 15.11269000745712 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.626770453407055 ms 13.671886747409213 ms 1.00

This comment was automatically generated by workflow using github-action-benchmark.

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

bigger_is_better

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.000491224608042 prompts/s 2.002675145423224 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4099.006519221878 tokens/s 4103.481372972186 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.912480139881211 prompts/s 0.9123163639924358 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.62241818455743 tokens/s 118.60112731901664 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.3518763324128744 prompts/s 2.35158613748706 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 726.3221281512932 tokens/s 726.2325082196705 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 539.7524824536448 tokens/s 539.6984248978303 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.82670730582109 prompts/s 13.888661076110504 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3553.46377759602 tokens/s 3569.3858965603995 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.21749763557733 prompts/s 2.2167695683344157 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 684.8224031365617 tokens/s 684.5975565568232 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 509.00145060690625 tokens/s 508.8165968661773 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.279846586244431 prompts/s 6.262518754196035 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 816.380056211776 tokens/s 814.1274380454846 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9834725792203591 prompts/s 0.983631103681614 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 289.9801682250536 tokens/s 290.0269096908695 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.27272149892232 tokens/s 212.30693741863956 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.1784135121874284 prompts/s 3.179495041356072 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1498.1960635857392 tokens/s 1498.7058596638462 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9839200949223194 prompts/s 0.9838806054260277 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.1121197214956 tokens/s 290.10047611188236 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.35619435343443 tokens/s 212.40342470138896 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.4740699333587688 prompts/s 0.47413571902199947 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.31564618405694 tokens/s 125.33303596627533 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.23863938310227 tokens/s 99.25241051527189 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.49188227177526656 prompts/s 0.4918873995766033 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 130.02415972107397 tokens/s 130.0255152040793 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.0195955517178 tokens/s 121.99791284298915 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.484731787411212 prompts/s 3.47554536374714 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1104.276656112739 tokens/s 1101.365570317831 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 796.6050402931531 tokens/s 794.5305234247786 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.070392337962599 prompts/s 7.079761182632859 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3305.832641537793 tokens/s 3310.213138551819 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.2402816701687509 prompts/s 0.24029023094156635 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.236617121937616 tokens/s 31.237730022403625 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.4870756874650213 prompts/s 3.4889400599782956 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3574.252579651647 tokens/s 3576.163561477753 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.892301184794405 prompts/s 13.862772772855124 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3570.321404492162 tokens/s 3562.732602623767 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.46381802710971487 prompts/s 0.4639583106048493 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 122.60565728618204 tokens/s 122.64273982528587 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 115.03923920393221 tokens/s 115.07403325161876 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.24050444846034622 prompts/s 0.2402768434229402 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.26557829984501 tokens/s 31.235989644982226 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.385294187765492 prompts/s 25.81640908488198 prompts/s 1.02
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1650.044122204757 tokens/s 1678.0665905173287 tokens/s 1.02
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.19079127527852 prompts/s 24.758579707575098 prompts/s 0.98
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1637.401432893104 tokens/s 1609.3076809923814 tokens/s 0.98
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0348366599974903 prompts/s 2.036458994186228 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4169.3803163348575 tokens/s 4172.704479087582 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.186567544846403 prompts/s 10.152973049137339 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1324.2537808300322 tokens/s 1319.886496387854 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 19.77455527601041 prompts/s 19.663083685272863 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2570.6921858813535 tokens/s 2556.2008790854725 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.7641779306994456 prompts/s 0.7639268340372671 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.34313099092793 tokens/s 99.31048842484472 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.9738424474381164 prompts/s 1.975317062191078 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 256.59951816695514 tokens/s 256.7912180848401 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.375070087786636 prompts/s 3.372912386549949 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.7591114122627 tokens/s 438.47861025149336 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.913582421623192 prompts/s 0.9127050864021167 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.76571481101496 tokens/s 118.65166123227517 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 16.805970074661225 prompts/s 16.74529333683377 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2184.7761097059592 tokens/s 2176.8881337883904 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9475005352302392 prompts/s 0.9475222154869255 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 279.37369114775345 tokens/s 279.38008364370495 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 204.53062386991687 tokens/s 204.5353038423927 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9128012486695777 prompts/s 0.9127306469948456 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.66416232704509 tokens/s 118.65498410932993 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.7264201651512283 prompts/s 2.7244396666690345 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 354.4346214696597 tokens/s 354.1771566669745 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 25.319411978814095 prompts/s 25.31894400415064 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1645.7617786229164 tokens/s 1645.7313602697916 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.042810991249061 prompts/s 4.044106530230551 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4143.881266030287 tokens/s 4145.209193486315 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.291204751027526 prompts/s 2.290440902922561 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 707.5851259106608 tokens/s 707.3492292465648 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 458.7877844060838 tokens/s 458.4974060258344 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.6986971480993076 prompts/s 3.6969976529082196 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 480.83062925291 tokens/s 480.6096948780686 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.44323094304649 prompts/s 11.384394975067053 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1487.6200225960436 tokens/s 1479.971346758717 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0350306388960138 prompts/s 2.0384319394042425 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4169.777779097932 tokens/s 4176.747043839293 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.124026058086653 prompts/s 4.119903332561249 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1306.8626175470793 tokens/s 1305.5561670553343 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 823.5844999041368 tokens/s 822.7144828880427 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.684509531093546 prompts/s 1.6806087990413068 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 549.3814464454329 tokens/s 548.1092720958677 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 382.89800046839554 tokens/s 381.9844535282326 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.366651970606848 prompts/s 3.3618719190890656 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1066.858342965604 tokens/s 1065.343592440134 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 769.5537963106074 tokens/s 768.4678895051088 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.7524535511334935 prompts/s 3.7569492477931656 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3846.264889911831 tokens/s 3850.8729789879944 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.451690096830156 prompts/s 7.459659949219531 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3822.71701967387 tokens/s 3826.8055539496195 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.168427151774745 prompts/s 10.14320166242732 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1321.8955297307168 tokens/s 1318.6162161155517 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.215235668692994 prompts/s 24.194691196838754 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3123.765401261396 tokens/s 3121.115164392199 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.936915637750351 prompts/s 1.938586235550944 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3968.740141750469 tokens/s 3972.163196643884 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.778916722675265 prompts/s 5.780744121662275 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2964.584278732411 tokens/s 2965.5217344127473 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.9299282069058705 prompts/s 2.926136100906273 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 380.89066689776314 tokens/s 380.3976931178155 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 14.15294931015704 prompts/s 14.13566061315486 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3637.307972710359 tokens/s 3632.864777580799 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.585753839345671 prompts/s 5.583551563685145 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2611.6750651244615 tokens/s 2610.645369116626 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9839561003649335 prompts/s 0.9839491630042055 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 290.1227360462685 tokens/s 290.12069054233336 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 212.40660338577823 tokens/s 212.39854615663114 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 15.849907964535024 prompts/s 15.85819651376316 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4073.4263468855015 tokens/s 4075.5565040371316 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.884539556368592 prompts/s 13.854177055679637 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3568.326665986728 tokens/s 3560.523503309667 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.752453117091788 prompts/s 4.748476807201662 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 617.8189052219325 tokens/s 617.3019849362161 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.4515583858947116 prompts/s 2.4510298349474677 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 757.1066044545775 tokens/s 756.9433738273766 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 562.903955357542 tokens/s 562.7597181437185 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7914316678740274 prompts/s 1.7931727190644227 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 232.88611682362355 tokens/s 233.11245347837496 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.7759490513155328 prompts/s 0.7735157651280619 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 100.87337667101926 tokens/s 100.55704946664804 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4.385737847551544 prompts/s 4.390290227894882 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1389.7964665106088 tokens/s 1391.239070317609 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1002.5358145718075 tokens/s 1003.5588820335794 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.8617249001990577 prompts/s 3.8650348835012154 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3958.268022704034 tokens/s 3961.660755588746 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 8.006897269328203 prompts/s 8.014925201315489 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4107.538299165369 tokens/s 4111.656628274846 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 28.75925609709302 prompts/s 28.794803850253647 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3709.944036525 tokens/s 3714.5296966827204 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.2246979903334574 prompts/s 2.2244860054437767 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 687.0460646947139 tokens/s 686.9805981078501 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 510.63937911721916 tokens/s 510.5996197455427 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.46584491800172917 prompts/s 0.465998128153441 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.97067709324092 tokens/s 126.01210716106583 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 118.04199658885149 tokens/s 118.08081901989425 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.24053724922113665 prompts/s 0.24030861859274258 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.269842398747766 tokens/s 31.240120417056538 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.069268476699994 prompts/s 2.0683856916680297 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4239.931108758287 tokens/s 4238.122282227793 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.8162848537281273 prompts/s 1.8186241599053863 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 577.089397431607 tokens/s 577.832666742845 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 406.67586560228096 tokens/s 407.20328597998866 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.299488794326514 prompts/s 13.304553986613938 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 864.4667716312234 tokens/s 864.796009129906 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 9.388754966244697 prompts/s 9.392961544017185 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2412.910026324887 tokens/s 2413.9911168124167 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7951249985191993 prompts/s 1.7943553898000457 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.3662498074959 tokens/s 233.26620067400594 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.453904477200255 prompts/s 7.4600015789981065 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3823.8529968037305 tokens/s 3826.980810026029 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.3722916447392928 prompts/s 3.3672515848879416 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.39791381610803 tokens/s 437.7427060354324 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.640128453997688 prompts/s 6.631018479858359 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 863.2166990196995 tokens/s 862.0324023815867 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.36975141985153 prompts/s 3.36660492732784 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 438.0676845806989 tokens/s 437.6586405526192 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.34813238123837 prompts/s 6.344786048806903 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 825.2572095609881 tokens/s 824.8221863448973 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.0138980660396932 prompts/s 1.0135982483583388 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 4\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 131.8067485851601 tokens/s 131.76777228658406 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.9680305401183158 prompts/s 0.9680309403748015 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 288.64411968401237 tokens/s 288.64423903115704 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 210.03681305793805 tokens/s 210.04012667272318 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.599282569516293 prompts/s 5.581164920707232 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 727.9067340371181 tokens/s 725.5514396919401 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.054054085829192 prompts/s 7.051461020667885 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 917.0270311577949 tokens/s 916.689932686825 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12.277898645005013 prompts/s 12.264694778738873 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1583.8489252056468 tokens/s 1582.1456264573146 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.566078530749362 prompts/s 11.50371467682034 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1503.590208997417 tokens/s 1495.4829079866442 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine throughput - 2:4 Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.587520707905341 prompts/s 5.578648697507994 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - 2:4 Sparse (with dataset)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2612.5011821882213 tokens/s 2608.3529850068376 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.219766283819688 prompts/s 24.22708159873138 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3124.34985061274 tokens/s 3125.293526236348 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7.452758547335559 prompts/s 7.460202709235207 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3823.265134783142 tokens/s 3827.083989837661 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.275917849188108 prompts/s 23.549919588363764 prompts/s 1.01
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3002.593402545266 tokens/s 3037.9396268989253 tokens/s 1.01
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1.7944179236028555 prompts/s 1.7930718518766309 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 8,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.2743300683712 tokens/s 233.09934074396202 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.4275292482464548 prompts/s 3.43365128693395 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1086.1497434768191 tokens/s 1088.0897563164995 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 783.4714906226711 tokens/s 784.884613075084 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.2562991848212249 prompts/s 0.2562593094265349 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 33.31889402675923 tokens/s 33.31371022544953 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.917265967113778 prompts/s 23.909559126416504 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3085.327309757677 tokens/s 3084.3331273077292 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.047291154220763 prompts/s 6.036791955001028 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2827.4714520674597 tokens/s 2822.5624464802804 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.73612307779246 prompts/s 25.284467324256294 prompts/s 1.02
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1607.8480000565098 tokens/s 1643.490376076659 tokens/s 1.02
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.420799994292652 prompts/s 11.415719977955193 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 64,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1484.7039992580449 tokens/s 1484.0435971341751 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - synthetic\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.80234884054723 prompts/s 3.803884961395343 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - synthetic\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 256,\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1460.1019547701362 tokens/s 1460.6918251758116 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.655513890317948 prompts/s 6.652228260231785 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - 2:4 Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 865.2168057413334 tokens/s 864.7896738301321 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5.586259392534818 prompts/s 5.583649217567208 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine throughput - Dense (with dataset)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"dataset\": \"sharegpt\",\n \"output-len\": 128,\n \"num-prompts\": 1000\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2611.91144157358 tokens/s 2610.691028165724 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.428011520641462 prompts/s 6.3452258756993265 prompts/s 0.99
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Dense (synthetic)\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 835.6414976833901 tokens/s 824.8793638409124 tokens/s 0.99
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.752991702000988 prompts/s 3.7554947864560835 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3846.816494551013 tokens/s 3849.3821561174855 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 0.49189708958399714 prompts/s 0.4918972664410856 prompts/s 1.00
{"name": "input_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 130.02807666063381 tokens/s 130.02812341103657 tokens/s 1.00
{"name": "output_throughput", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.99375753076185 tokens/s 121.99052207738923 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2.0345133750303592 prompts/s 2.037665339068763 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2048,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4168.717905437205 tokens/s 4175.176279751895 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 48.830049111515606 prompts/s 48.79581921111015 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 64,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3173.9531922485144 tokens/s 3171.72824872216 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 8.207179680238195 prompts/s 8.19876334119261 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 512,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 4210.283175962194 tokens/s 4205.965594031809 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3.7536127537480835 prompts/s 3.752344427261337 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine prefill throughput - Dense (synthetic)\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 1024,\n \"output-len\": 1,\n \"num-prompts\": 1\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3847.4530725917857 tokens/s 3846.1530379428705 tokens/s 1.00
{"name": "request_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6.641418649529541 prompts/s 6.633908223317837 prompts/s 1.00
{"name": "token_throughput", "description": "VLLM Engine decode throughput - Sparse (synthetic)\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax_model_len - 4096\nbenchmark_throughput {\n \"use-all-available-gpus_\": \"\",\n \"input-len\": 2,\n \"output-len\": 128,\n \"num-prompts\": 32,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 863.3844244388403 tokens/s 862.4080690313189 tokens/s 1.00

This comment was automatically generated by workflow using github-action-benchmark.

@github-actions
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

smaller_is_better

Benchmark suite Current: df1f1a0 Previous: e8e00d2 Ratio
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5376.062333001755 ms 5385.054345500976 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 184.5827273187315 ms 183.15599952013872 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 173.52721449969977 ms 174.65145149981254 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.43196687663791 ms 40.29276032413655 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.58095798642143 ms 37.395030997344755 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12320.661744000063 ms 12239.373073500246 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 200.01820418526768 ms 196.4258269800266 ms 1.02
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 169.44416600108525 ms 172.22716749984102 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 96.04773220335846 ms 95.72504978387069 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 101.17233322299779 ms 100.80430680167527 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 3652.386217499952 ms 3651.734495499113 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 142.93804023662233 ms 143.39316453994493 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 95.80902800007607 ms 94.5331505008653 ms 1.01
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 24.498444220468098 ms 24.4685511837163 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 23.784562093061254 ms 23.80589036230062 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1823.0353210010435 ms 1813.6266555002294 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 93.06960186000651 ms 93.09651339993555 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 59.665249999852676 ms 60.07437150037731 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 12.502380206719828 ms 12.492343204249421 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.783106806446016 ms 11.77166285193115 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 5238.431358498929 ms 5235.645166498216 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 91.44519897992723 ms 91.47958961346981 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 60.88910249854962 ms 61.74462950002635 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 37.252881617494005 ms 37.26582722049259 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 31.70269005355473 ms 31.708689124541355 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 1929.8616509995554 ms 1925.8178315003533 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 79.59311796000596 ms 79.36294155328748 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.459064500078966 ms 37.76919200026896 ms 1.04
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 10.979780574097273 ms 10.98752899201461 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.09183504894853 ms 11.090169800305919 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 65573.8574650004 ms 66556.05931600257 ms 0.99
{"name": "mean_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 28950.78831299604 ms 29395.456832029286 ms 0.98
{"name": "median_ttft_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 28142.69032300217 ms 28430.397764501322 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 188.86090202522692 ms 188.6404633420125 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned50\nmax-model-len - 4096\nsparsity - sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 199.33138919321786 ms 200.07395705000116 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6378.783236000345 ms 6367.810279999503 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 108.38548640673253 ms 111.16892662006042 ms 0.97
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 71.50023299982422 ms 69.513698499577 ms 1.03
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 36.32068798008349 ms 36.28406362460848 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 36.99483870421225 ms 36.9687584896116 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6047.278685499805 ms 6061.940364000293 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 125.4039507166696 ms 128.1418428899633 ms 0.98
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 87.60409249862278 ms 87.2189094998248 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.662599829984984 ms 40.458295919317536 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40.43363331252289 ms 40.42338318231981 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6089.519108001696 ms 6064.388077998956 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 120.69910238930122 ms 122.99509105597099 ms 0.98
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 89.65837299911072 ms 91.56941200126312 ms 0.98
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 55.06700430067017 ms 55.034764851352534 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 48.04969955440588 ms 48.14404020383918 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 18383.634987500045 ms 18630.98225700014 ms 0.99
{"name": "mean_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 977.7821168053584 ms 1066.207715544699 ms 0.92
{"name": "median_ttft_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 233.23141549917636 ms 232.56485849924502 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 161.4755453013489 ms 162.23560618336472 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - 2:4 Sparse\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-pruned2.4\nmax-model-len - 4096\nsparsity - semi_structured_sparse_w16a16\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 147.98108700900505 ms 149.11634172076177 ms 0.99
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 78107.6898225001 ms 78389.37180349967 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 58225.91374190133 ms 58591.156928930686 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 71513.63180799945 ms 72002.09738900048 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 67.80498533429089 ms 68.32876132892734 ms 0.99
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 65.44014757829031 ms 65.84333552122904 ms 0.99
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 66338.3866639997 ms 66787.84569050003 ms 0.99
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 22944.870291643343 ms 23200.36978493534 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 17367.390752000574 ms 17407.502998499695 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 260.65817349493784 ms 261.4966492002461 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 277.99167895448284 ms 279.02716963159884 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2134.466259499277 ms 2124.492699500479 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 136.62190305008457 ms 136.58727046664958 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 102.67750049933966 ms 102.26584150041163 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 15.099409057440004 ms 15.11269000745712 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 13.654440891257662 ms 13.671886747409213 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2408.684269999867 ms 2411.166190499898 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 116.16479182931411 ms 116.23203609204211 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 80.27312500053085 ms 80.91534550021606 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 18.106101958196053 ms 18.117534204784953 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 16.139953761015917 ms 16.20924310544079 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 14822.945272499965 ms 14700.193791500169 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 205.64598047202162 ms 208.177746926672 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 163.80199600098422 ms 164.7100544996647 ms 0.99
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 119.56480416917998 ms 118.11084162295658 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - neuralmagic/OpenHermes-2.5-Mistral-7B-marlin\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 126.0499936903422 ms 125.06230301204924 ms 1.01
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 7886.070769500293 ms 7866.947419000098 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 157.57009318131654 ms 156.33848048406313 ms 1.01
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 132.7795765000701 ms 125.77558449993376 ms 1.06
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 58.182222314526086 ms 57.926682457080105 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 57.73655366013833 ms 57.515965234391565 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6141.3096454998595 ms 6157.483541000147 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 104.48978885987647 ms 105.50076459997702 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 67.92814349955734 ms 63.466036500358314 ms 1.07
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 34.73078370484817 ms 34.73502434048436 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 35.22087641082219 ms 35.23139902220215 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 259394.88207350043 ms 257960.56549199967 ms 1.01
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 244259.34381698666 ms 243138.74433172468 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 245936.9741119999 ms 244561.7894119996 ms 1.01
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 70.70314073132613 ms 70.2651706463995 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 66.68409539982376 ms 66.51669527142957 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 6071.920400999261 ms 6079.410813499635 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 121.85457794001802 ms 122.25182649330236 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 78.61482199950842 ms 83.33002849940385 ms 0.94
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.801395855077615 ms 39.852174955426804 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - NousResearch/Llama-2-7b-chat-hf\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39.502406777722285 ms 39.52926163180674 ms 1.00
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 63004.24120249954 ms 64089.48246949967 ms 0.98
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 40527.186443927385 ms 41101.79994840606 ms 0.99
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 39565.95233650023 ms 40744.35401400115 ms 0.97
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 99.86915523887204 ms 99.1521024514054 ms 1.01
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - teknium/OpenHermes-2.5-Mistral-7B\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 102.12643246719269 ms 100.95686430257042 ms 1.01
{"name": "median_request_latency", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 2066.5399399995295 ms 2067.7271800004746 ms 1.00
{"name": "mean_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 116.18510289994447 ms 116.29355367333726 ms 1.00
{"name": "median_ttft_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 86.63544850060134 ms 86.71847000005073 ms 1.00
{"name": "mean_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.6430866408856 ms 11.620019052349596 ms 1.00
{"name": "median_tpot_ms", "description": "VLLM Serving - Dense\nmodel - TheBloke/OpenHermes-2.5-Mistral-7B-GPTQ\nmax-model-len - 4096\nsparsity - None\nbenchmark_serving {\n \"nr-qps-pair_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.2.0", "python_version": "3.10.12 (main, Mar 7 2024, 18:39:53) [GCC 9.4.0]", "torch_version": "2.2.1+cu121"} 11.652037467004137 ms 11.619994200319992 ms 1.00

This comment was automatically generated by workflow using github-action-benchmark.

Please sign in to comment.