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This repository has been archived by the owner on Oct 11, 2024.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

Commit

Permalink
requirements-dev.txt and workflow patches (#255)
Browse files Browse the repository at this point in the history 
SUMMARY:
* update `requirements-dev.txt` to address `urllib` dependency
* update "build test" only be reusable
* update "benchmark" workflow to disambiguate artifacts

TEST PLAN:
runs on remote push

---------

Co-authored-by: Michael Goin <[email protected]>
Co-authored-by: dhuangnm <[email protected]>
Co-authored-by: andy-neuma <[email protected]>
Co-authored-by: Domenic Barbuzzi <[email protected]>
  • Loading branch information
@andy-neuma @mgoin
@dhuangnm @dbarbuzzi
5 people authored May 22, 2024
1 parent 93183d6 commit a10b831
Show file tree
Hide file tree
Showing 11 changed files with 117 additions and 65 deletions.
12 changes: 6 additions & 6 deletions .github/actions/nm-copy-benchmark-data-to-efs/action.yml
View file
Original file line number Diff line number Diff line change
@@ -1,33 +1,33 @@
name: copy benchmark data to EFS
name: copy benchmark data to EFS
description: "Copies the given input folder to EFS under a well-defined directory structure.
The directory structure is as follows,
date/github_event/label/branch/commit-hash/github_run_id/.
The directory structure is created if it doesn't exist"
inputs:
label:
description: "requested runner label (specifies instance)"
type: string
required: true
python:
description: 'python version, e.g. 3.10.12'
required: true
src:
description: "Src benchmark folder to copy"
type: string
required: true
efs_dst:
description: "Destination EFS path to copy the src folder to"
type: string
required: true

runs:
using: composite
steps:
- id: copy_benchmark_data_to_efs
- id: copy_benchmark_data_to_efs
run: |
echo "event name ${{ github.event_name }}"
echo "ref ${{ github.ref }}"
echo "sha ${{ github.sha }}"
echo "run ${{ github.run_id }}"
SUCCESS=0
./.github/scripts/nm-store-benchmark-data.sh -i ${{ inputs.src }} -o ${{ inputs.efs_dst }} -l ${{ inputs.label }} -e ${{ github.event_name }} -b ${{ github.ref }} -c ${{ github.sha }} -r ${{ github.run_id }} || SUCCESS=$?
./.github/scripts/nm-store-benchmark-data.sh -i ${{ inputs.src }} -o ${{ inputs.efs_dst }} -l ${{ inputs.label }} -p ${{ inputs.python }} -e ${{ github.event_name }} -b ${{ github.ref }} -c ${{ github.sha }} -r ${{ github.run_id }} || SUCCESS=$?
echo "test=${SUCCESS}" >> "$GITHUB_OUTPUT"
exit ${SUCCESS}
shell: bash
14 changes: 10 additions & 4 deletions .github/scripts/nm-store-benchmark-data.sh
View file
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@ usage() {
echo " -o - path to the destination-root"
echo " -e - github event name"
echo " -l - github instance label"
echo " -p - python version"
echo " -b - github branch name"
echo " -c - github commit hash"
echo " -r - github run id"
Expand All @@ -28,11 +29,12 @@ INPUT_PATH=""
OUTPUT_PATH=""
GITHUB_EVENT_NAME=""
GITHUB_LABEL=""
PYTHON_VERSION=""
GITHUB_BRANCH=""
GITHUB_COMMIT=""
GITHUB_RUN_ID=""

while getopts "hi:o:e:l:b:c:r:" OPT; do
while getopts "hi:o:e:l:p:b:c:r:" OPT; do
case "${OPT}" in
h)
usage
Expand All @@ -50,6 +52,9 @@ while getopts "hi:o:e:l:b:c:r:" OPT; do
l)
GITHUB_LABEL="${OPTARG}"
;;
p)
PYTHON_VERSION="${OPTARG}"
;;
b)
GITHUB_BRANCH="${OPTARG}"
;;
Expand All @@ -68,12 +73,13 @@ echo "INPUT_PATH : ${INPUT_PATH}"
echo "OUTPUT_PATH : ${OUTPUT_PATH}"
echo "GITHUB_EVENT_NAME : ${GITHUB_EVENT_NAME}"
echo "GITHUB_LABEL : ${GITHUB_LABEL}"
echo "PYTHON VERSION: ${PYTHON_VERSION}"
echo "GITHUB_BRANCH : ${GITHUB_BRANCH}"
echo "GITHUB_COMMIT : ${GITHUB_COMMIT}"
echo "GITHUB_RUN_ID : ${GITHUB_RUN_ID}"

# Make sure we have all the information to construct a correct path
if [[ "${INPUT_PATH}" == "" || "${OUTPUT_PATH}" == "" || "${GITHUB_EVENT_NAME}" == "" || "${GITHUB_LABEL}" == "" || "${GITHUB_BRANCH}" == "" || "${GITHUB_COMMIT}" == "" || "${GITHUB_RUN_ID}" == "" ]];
if [[ "${INPUT_PATH}" == "" || "${OUTPUT_PATH}" == "" || "${GITHUB_EVENT_NAME}" == "" || "${GITHUB_LABEL}" == "" || "${PYTHON_VERSION}" == "" || "${GITHUB_BRANCH}" == "" || "${GITHUB_COMMIT}" == "" || "${GITHUB_RUN_ID}" == "" ]];
then
echo "Error : Incomplete arg list - Atleast one of the arguments is an empty string"
exit 1
Expand All @@ -86,7 +92,7 @@ GITHUB_COMMIT=${GITHUB_COMMIT:0:7}
# Get today's date
TODAY=`date '+%Y-%m-%d'`

DESTINATION_DIR=${OUTPUT_PATH}/${TODAY}/${GITHUB_EVENT_NAME}/${GITHUB_LABEL}/${GITHUB_BRANCH}/${GITHUB_COMMIT}/${GITHUB_RUN_ID}
DESTINATION_DIR=${OUTPUT_PATH}/${TODAY}/${GITHUB_EVENT_NAME}/${GITHUB_LABEL}/${PYTHON_VERSION}/${GITHUB_BRANCH}/${GITHUB_COMMIT}/${GITHUB_RUN_ID}
echo "Destination DIR : ${DESTINATION_DIR}"

# Create destination dir
Expand All @@ -102,6 +108,6 @@ then
exit 1
fi
# Tar file
tar -cvf ${DESTINATION_TAR} ${INPUT_PATH}
tar -cvf ${DESTINATION_TAR} ${INPUT_PATH}

exit 0
44 changes: 0 additions & 44 deletions .github/workflows/build-test.yml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -67,50 +67,6 @@ on:
type: string
default: "false"

# makes workflow manually callable
workflow_dispatch:
inputs:
build_label:
description: "requested runner label (specifies instance)"
type: string
required: true
build_timeout:
description: "time limit for build in minutes "
type: string
required: true
test_label_solo:
description: "requested runner label (specifies instance)"
type: string
required: true
test_label_multi:
description: "requested runner label (specifies instance)"
type: string
required: true
test_timeout:
description: "time limit for test run in minutes "
type: string
required: true
gitref:
description: "git commit hash or branch name"
type: string
required: true
Gi_per_thread:
description: 'requested GiB to reserve per thread'
type: string
required: true
nvcc_threads:
description: "number of threads nvcc build threads"
type: string
required: true
python:
description: "python version, e.g. 3.10.12"
type: string
required: true
test_skip_list:
description: 'file containing tests to skip'
type: string
required: true

jobs:

BUILD:
Expand Down
6 changes: 4 additions & 2 deletions .github/workflows/nm-benchmark.yml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -135,7 +135,7 @@ jobs:
if: success()
uses: actions/upload-artifact@v4
with:
name: ${{ github.run_id }}-${{ inputs.label }}
name: ${{ github.run_id }}-${{ inputs.label }}-${{ inputs.python }}
path: benchmark-results
retention-days: 2

Expand All @@ -144,6 +144,7 @@ jobs:
uses: ./.github/actions/nm-copy-benchmark-data-to-efs
with:
label: ${{ inputs.label }}
python: ${{ inputs.python }}
src: benchmark-results
efs_dst: /EFS/benchmark_results

Expand All @@ -164,7 +165,7 @@ jobs:
- name: set gh action benchmark input artifact name
id: set_gh_action_benchmark_input_artifact_name
run: |
GH_ACTION_BENCHMARK_INPUT_ARTIFACT_NAME=`echo "gh_action_benchmark_jsons-${{ github.run_id }}-${{ inputs.label }}"`
GH_ACTION_BENCHMARK_INPUT_ARTIFACT_NAME=`echo "gh_action_benchmark_jsons-${{ github.run_id }}-${{ inputs.label }}-${{ inputs.python }}"`
echo "gh_action_benchmark_input_artifact_name=$GH_ACTION_BENCHMARK_INPUT_ARTIFACT_NAME" >> $GITHUB_OUTPUT
- name: store gh action benchmark input artifacts
Expand All @@ -180,6 +181,7 @@ jobs:
uses: ./.github/actions/nm-copy-benchmark-data-to-efs
with:
label: ${{ inputs.label }}
python: ${{ inputs.python }}
src: gh-action-benchmark-jsons
efs_dst: /EFS/benchmark_results

Expand Down
20 changes: 18 additions & 2 deletions .github/workflows/remote-push.yml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,23 @@ concurrency:

jobs:

BUILD-TEST:
BUILD-TEST-3-8:
uses: ./.github/workflows/build-test.yml
with:
python: 3.8.17
gitref: ${{ github.ref }}

test_label_solo: aws-avx2-32G-a10g-24G
test_label_multi: ignore
test_timeout: 480
test_skip_list: neuralmagic/tests/skip-for-remote-push.txt

benchmark_label: aws-avx2-32G-a10g-24G
benchmark_config_list_file: ./.github/data/nm_benchmark_remote_push_configs_list.txt
benchmark_timeout: 480
secrets: inherit

BUILD-TEST-3-10:
uses: ./.github/workflows/build-test.yml
with:
python: 3.10.12
Expand All @@ -24,5 +40,5 @@ jobs:

benchmark_label: aws-avx2-32G-a10g-24G
benchmark_config_list_file: ./.github/data/nm_benchmark_remote_push_configs_list.txt
benchmark_timeout: 180
benchmark_timeout: 480
secrets: inherit
53 changes: 53 additions & 0 deletions neuralmagic/tests/skip-for-remote-push-tmp.txt
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,53 @@
tests/test_sequence.py
tests/metrics/test_metrics.py
tests/kernels/test_prefix_prefill.py
tests/kernels/test_pos_encoding.py
tests/kernels/test_activation.py
tests/kernels/test_moe.py
tests/kernels/test_layernorm.py
tests/kernels/test_attention.py
tests/core/test_block_manager.py
tests/distributed/test_basic_distributed_correctness.py
tests/distributed/test_chunked_prefill_distributed.py
tests/distributed/test_comm_ops.py
tests/distributed/test_custom_all_reduce.py
tests/distributed/test_pynccl_library.py
tests/distributed/test_pynccl.py
tests/prefix_caching/test_prefix_caching.py
tests/models/test_models_logprobs.py
tests/models/test_models.py
tests/spec_decode/test_utils.py
tests/spec_decode/test_multi_step_worker.py
tests/spec_decode/test_spec_decode_worker.py
tests/spec_decode/test_batch_expansion.py
tests/spec_decode/test_metrics.py
tests/spec_decode/test_ngram_worker.py
tests/spec_decode/e2e/test_logprobs.py
tests/spec_decode/e2e/test_ngram_correctness.py
tests/spec_decode/e2e/test_compatibility.py
tests/spec_decode/e2e/test_multistep_correctness.py
tests/spec_decode/test_metrics.py
tests/test_sampling_params.py
tests/async_engine/test_async_llm_engine.py
tests/async_engine/test_chat_template.py
tests/async_engine/test_request_tracker.py
tests/samplers/test_logprobs.py
tests/samplers/test_seeded_generate.py
tests/samplers/test_rejection_sampler.py
tests/samplers/test_sampler.py
tests/entrypoints/test_guided_processors.py
tests/entrypoints/test_openai_server.py
tests/lora/test_utils.py
tests/lora/test_tokenizer.py
tests/lora/test_layer_variation.py
tests/lora/test_gemma.py
tests/lora/test_lora_manager.py
tests/lora/test_worker.py
tests/lora/test_mixtral.py
tests/lora/test_punica.py
tests/lora/test_lora.py
tests/worker/test_model_runner.py
tests/engine/test_detokenize.py
tests/engine/test_computed_prefix_blocks.py
tests/accuracy/test_lm_eval_correctness.py
tests/tensorizer_loader/test_tensorizer.py
4 changes: 2 additions & 2 deletions requirements-dev.txt
View file
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,7 @@ isort==5.13.2
# type checking
mypy==1.9.0
types-PyYAML
types-requests
types-requests==2.31.0.2
types-setuptools

# testing
Expand All @@ -21,7 +21,7 @@ pytest-rerunfailures
pytest-shard
httpx
einops # required for MPT
requests
requests==2.31
ray
peft
awscli
Expand Down
7 changes: 7 additions & 0 deletions tests/engine/test_multiproc_workers.py
View file
Original file line number Diff line number Diff line change
@@ -1,4 +1,6 @@
import asyncio
# UPSTREAM SYNC
import sys
from concurrent.futures import ThreadPoolExecutor
from functools import partial
from time import sleep
Expand Down Expand Up @@ -100,6 +102,11 @@ def execute_workers(worker_input: str) -> None:
def test_local_workers_clean_shutdown() -> None:
"""Test clean shutdown"""

# UPSTREAM SYNC
pytest.mark.skipif(sys.version_info < (3, 10),
reason="This test is inexplicably failing in CI "
"on Python < 3.10")

workers, worker_monitor = _start_workers()

assert worker_monitor.is_alive()
Expand Down
12 changes: 12 additions & 0 deletions tests/models/test_big_models.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,9 @@
Run `pytest tests/models/test_big_models.py`.
"""
# UPSTREAM SYNC
import sys

import pytest

MODELS = [
Expand All @@ -27,6 +30,11 @@
"EleutherAI/gpt-j-6b",
]

# UPSTREAM SYNC
SKIPPED_MODELS_PY38 = [
"mosaicml/mpt-7b",
]


@pytest.mark.parametrize("model", MODELS)
@pytest.mark.parametrize("dtype", ["half"])
Expand All @@ -45,6 +53,10 @@ def test_models(
if model in SKIPPED_MODELS_OOM:
pytest.skip(reason="These models cause OOM issue on the CPU"
"because it is a fp32 checkpoint.")
# UPSTREAM SYNC
if model in SKIPPED_MODELS_PY38 and sys.version_info < (3, 9):
pytest.skip(reason="This model has custom code that does not "
"support Python 3.8")

hf_model = hf_runner(model, dtype=dtype)
hf_outputs = hf_model.generate_greedy(example_prompts, max_tokens)
Expand Down
6 changes: 3 additions & 3 deletions vllm/engine/async_llm_engine.py
View file
Original file line number Diff line number Diff line change
@@ -1,8 +1,8 @@
import asyncio
import time
from functools import partial
from typing import (Any, AsyncIterator, Callable, Dict, Iterable, List,
Optional, Set, Tuple, Type, Union)
from typing import (AsyncIterator, Callable, Dict, Iterable, List, Optional,
Set, Tuple, Type, Union)

from transformers import PreTrainedTokenizer

Expand Down Expand Up @@ -327,7 +327,7 @@ def __init__(self,
# We need to keep a reference to unshielded
# task as well to prevent it from being garbage
# collected
self._background_loop_unshielded: Optional[asyncio.Task[Any]] = None
self._background_loop_unshielded: Optional[asyncio.Task] = None
self.start_engine_loop = start_engine_loop
self._errored_with: Optional[BaseException] = None

Expand Down
4 changes: 2 additions & 2 deletions vllm/entrypoints/openai/api_server.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,7 @@
import re
from contextlib import asynccontextmanager
from http import HTTPStatus
from typing import Any, Set
from typing import Set

import fastapi
import uvicorn
Expand Down Expand Up @@ -34,7 +34,7 @@
openai_serving_completion: OpenAIServingCompletion
logger = init_logger(__name__)

_running_tasks: Set[asyncio.Task[Any]] = set()
_running_tasks: Set[asyncio.Task] = set()


@asynccontextmanager
Expand Down

2 comments on commit a10b831

@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: a10b831 Previous: 93183d6 Ratio
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 5.653370152875044 prompts/s 5.659332130139428 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2643.2897486782554 tokens/s 2646.077330767991 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1.803163981319349 prompts/s 1.80231469550631 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 234.41131757151535 tokens/s 234.3009104158203 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.2405614105575201 prompts/s 0.24077447017190282 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 31.27298337247761 tokens/s 31.300681122347363 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 8.097699325321036 prompts/s 8.029085716271409 prompts/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4154.119753889691 tokens/s 4118.920972447233 tokens/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.2289266095739215 prompts/s 2.228605355011266 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 688.3519750793489 tokens/s 688.2527631036126 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 511.6575329830589 tokens/s 511.59567382257956 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 7.895107558899879 prompts/s 7.83560364474303 prompts/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4050.1901777156377 tokens/s 4019.6646697531746 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 5.659511090871358 prompts/s 5.656699649673993 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2646.1610056478116 tokens/s 2644.8464882015724 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.464061618449088 prompts/s 0.4640638720564649 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 122.67004822083193 tokens/s 122.67064393940593 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 115.0996563518658 tokens/s 115.10021530659147 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 26.073381813489554 prompts/s 26.12036651613379 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1694.769817876821 tokens/s 1697.8238235486963 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 8.047647351548099 prompts/s 8.032491298797158 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4128.443091344175 tokens/s 4120.668036282942 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.045333441250376 prompts/s 2.043094072044478 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4190.888221122021 tokens/s 4186.299753619135 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4.440815106995149 prompts/s 4.427865441643753 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1407.249899255693 tokens/s 1403.146279802489 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1015.1466491118539 tokens/s 1012.2041361391732 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.914893409909721 prompts/s 0.9140012862778002 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 118.93614328826372 tokens/s 118.82016721611403 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.9258180213767697 prompts/s 2.92796304913923 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 380.3563427789801 tokens/s 380.6351963880999 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.072702975220488 prompts/s 2.0726570638451607 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4246.96839622678 tokens/s 4246.874323818734 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 17.517682267739723 prompts/s 17.50036947518833 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2277.298694806164 tokens/s 2275.0480317744828 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 26.045830608621227 prompts/s 26.10486627637359 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1692.9789895603797 tokens/s 1696.8163079642832 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4.38260347800527 prompts/s 4.366970072688601 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1388.80321614509 tokens/s 1383.8491463342907 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 998.9823237191291 tokens/s 998.0622761728343 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 5.660722262217961 prompts/s 5.654522326493009 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2646.7273009226296 tokens/s 2643.828458975071 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 9.883575288284758 prompts/s 9.857012749936569 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1284.8647874770186 tokens/s 1281.411657491754 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3.4887785423196624 prompts/s 3.477063615124994 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1105.5590322756777 tokens/s 1101.8466889969593 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 797.4812801699593 tokens/s 794.8080635269617 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1.695219434084152 prompts/s 1.6941532394054397 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 552.8743456470478 tokens/s 552.5266197577183 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 385.3075553192763 tokens/s 385.0516659940401 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 11.569508455544167 prompts/s 11.60129898649921 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1504.0360992207418 tokens/s 1508.168868244897 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 16.846767987465164 prompts/s 16.89814696257658 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2190.0798383704714 tokens/s 2196.7591051349555 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.451987809666257 prompts/s 2.452049884597038 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 757.2392219665313 tokens/s 757.2583923604878 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 563.113712416714 tokens/s 563.347018086708 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 20.536385060253096 prompts/s 20.613890132035255 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2669.7300578329023 tokens/s 2679.8057171645833 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 11.563733859211244 prompts/s 11.57089513929368 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1503.2854016974616 tokens/s 1504.2163681081784 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 24.453384928470584 prompts/s 24.47284392939348 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3154.486655772705 tokens/s 3156.996866891759 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 29.08409558622885 prompts/s 28.828762341938397 prompts/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3751.848330623522 tokens/s 3718.910342110053 tokens/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.4750630316952714 prompts/s 0.47497397224660304 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 125.57816179832804 tokens/s 125.55461982366705 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 99.44652796821015 tokens/s 99.42788485695557 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 10.402638789531832 prompts/s 10.41308257991421 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1352.343042639138 tokens/s 1353.7007353888473 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3.4013805439594083 prompts/s 3.398949532170632 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 442.1794707147231 tokens/s 441.86343918218216 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1.8015286616542001 prompts/s 1.8019104111398394 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 234.19872601504602 tokens/s 234.24835344817913 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 14.250346960214298 prompts/s 14.25870646232627 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3662.3391687750745 tokens/s 3664.4875608178513 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 10.288920640428751 prompts/s 10.296932575494488 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1337.5596832557376 tokens/s 1338.6012348142833 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.4919038913234492 prompts/s 0.4919001039533992 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 130.02987463244057 tokens/s 130.02887347904155 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 121.88066683284875 tokens/s 121.99778444849572 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.0428822850463124 prompts/s 2.0421154745560894 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4185.865802059894 tokens/s 4184.2946073654275 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 5.37763694149466 prompts/s 5.380014300689102 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 699.0928023943059 tokens/s 699.4018590895832 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4.167070855161931 prompts/s 4.134864664470034 prompts/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4271.24762654098 tokens/s 4238.236281081785 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\": 128,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 22.796088418330164 prompts/s 23.50286955763114 prompts/s 1.03
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2940.6954059645914 tokens/s 3031.870172934417 tokens/s 1.03
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.2405388627227632 prompts/s 0.2405921272511396 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 31.270052153959217 tokens/s 31.276976542648146 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.46605215786316073 prompts/s 0.4661118238497531 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 126.02671751497016 tokens/s 126.04285199329122 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 118.09450978813916 tokens/s 118.10962875136842 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.355711809219915 prompts/s 2.3555280996495567 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 727.5066256686889 tokens/s 727.4498912544404 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 540.6358602159704 tokens/s 540.5905581654404 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.00847747007683 prompts/s 1.9784295897435635 prompts/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4115.370336187425 tokens/s 4053.8022293845615 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.7220531207903558 prompts/s 0.7236921443345576 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 93.86690570274625 tokens/s 94.0799787634925 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.9837515545990728 prompts/s 0.9838593350399324 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 290.0624250453853 tokens/s 290.0942044676409 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 212.33293554466388 tokens/s 212.35619887501903 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3.2029517133801586 prompts/s 3.2040326229370235 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1509.7625373291517 tokens/s 1510.272041343333 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 20.16576913203148 prompts/s 20.20583735768229 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2621.5499871640923 tokens/s 2626.7588564986977 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 7.501095985216715 prompts/s 7.492423717246277 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3848.0622404161745 tokens/s 3843.61336694734 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 6.734711568624295 prompts/s 6.746325167421155 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 875.5125039211583 tokens/s 877.0222717647501 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 13.937727654623218 prompts/s 13.85215393834894 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\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3581.996007238167 tokens/s 3560.003562155677 tokens/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.301628054155174 prompts/s 2.301977032008458 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 710.8041198712286 tokens/s 710.9118935383988 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 460.65090759602214 tokens/s 460.89877204602413 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.757319192749701 prompts/s 0.7576141898514492 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 98.45149505746114 tokens/s 98.4898446806884 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3.7684033937001664 prompts/s 3.766050232897849 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3862.6134785426707 tokens/s 3860.2014887202954 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1.8009478904716558 prompts/s 1.8023684875628179 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 234.12322576131524 tokens/s 234.30790338316632 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.4473115563986405 prompts/s 2.446351457478283 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 755.7950702574041 tokens/s 755.4985661081599 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 561.3643248067201 tokens/s 561.254998582441 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4.173728717646616 prompts/s 4.1719723802362205 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1322.6128933350362 tokens/s 1322.0563275730558 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 833.3405881943821 tokens/s 832.8619715262373 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.24049576385492452 prompts/s 0.24051294748537014 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 31.264449301140186 tokens/s 31.266683173098116 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3.398025761897617 prompts/s 3.3986233396532435 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 441.7433490466902 tokens/s 441.8210341549216 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.9839342996918764 prompts/s 0.9839106260395385 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 290.11630804514874 tokens/s 290.1093277898447 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 212.3100634065141 tokens/s 212.39678684315516 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 5.621720871736365 prompts/s 5.641638691506198 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 730.8237133257275 tokens/s 733.4130298958057 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 13.931874915076238 prompts/s 13.916032108974909 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3580.491853174593 tokens/s 3576.4202520065514 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 26.046203047543937 prompts/s 25.45289222104943 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1693.003198090356 tokens/s 1654.4379943682131 tokens/s 0.98
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 6.744936344113242 prompts/s 6.735641625328046 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 876.8417247347214 tokens/s 875.633411292646 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 15.945378785586424 prompts/s 15.783074395706146 prompts/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4097.962347895711 tokens/s 4056.2501196964795 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3.871861337568351 prompts/s 3.87102762547275 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3968.6578710075596 tokens/s 3967.8033161095686 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1.8321741008562966 prompts/s 1.8237804236354633 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 582.1378985144717 tokens/s 579.4709698553115 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 410.2372171251311 tokens/s 408.3626746562166 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.9477220467836512 prompts/s 0.9474953431034917 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 279.43900456764885 tokens/s 279.3721602318749 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 204.57528101871895 tokens/s 204.52950308013007 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3.8357435236452475 prompts/s 3.836080913807782 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1472.925513079775 tokens/s 1473.0550709021882 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4.054683998514079 prompts/s 4.000491116300599 prompts/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4156.051098476932 tokens/s 4100.503394208114 tokens/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 7.049972744358004 prompts/s 6.984992744729419 prompts/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3296.2852563520282 tokens/s 3265.903207725687 tokens/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 6.453279591484572 prompts/s 6.454478505595475 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 838.9263468929943 tokens/s 839.0822057274117 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.063691708954948 prompts/s 2.0321267699844623 prompts/s 0.98
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4228.504311648689 tokens/s 4163.8277516981625 tokens/s 0.98
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 28.558117222684473 prompts/s 28.361800827803343 prompts/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3683.997121726297 tokens/s 3658.6723067866315 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\": 16\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3.4005113438590664 prompts/s 3.397721782099698 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 442.0664747016786 tokens/s 441.70383167296075 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 7.500426418069211 prompts/s 7.492978666973118 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3847.7187524695055 tokens/s 3843.89805615721 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 24.4985075863086 prompts/s 24.46617078568386 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3160.3074786338093 tokens/s 3156.1360313532177 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.9141678455580275 prompts/s 0.8960533371090144 prompts/s 0.98
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 118.84181992254359 tokens/s 116.48693382417186 tokens/s 0.98
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1.9857535581074708 prompts/s 1.9859788922041888 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 258.1479625539712 tokens/s 258.17725598654454 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 25.511984063479975 prompts/s 25.543770027463562 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1658.2789641261984 tokens/s 1660.3450517851315 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 49.475282464728785 prompts/s 49.09811910905685 prompts/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3215.893360207371 tokens/s 3191.3777420886954 tokens/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.9680068355662015 prompts/s 0.9680208233603937 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 288.63705153967805 tokens/s 288.64122237352433 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 210.0348964917025 tokens/s 210.0347047805825 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 47.95810226294661 prompts/s 47.07098888881206 prompts/s 0.98
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3117.27664709153 tokens/s 3059.614277772784 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.044326100667966 prompts/s 2.0423510658262263 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4188.8241802686625 tokens/s 4184.777333877937 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2.7935917672727664 prompts/s 2.7927030163850133 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 363.16692974545964 tokens/s 363.05139213005174 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.9839630570774264 prompts/s 0.9839319918123051 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 290.12478725613613 tokens/s 290.1156275591642 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 212.5720989741503 tokens/s 212.41123839244042 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 11.579108965741545 prompts/s 11.569277859797676 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1505.2841655464008 tokens/s 1504.006121773698 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3.726723671509959 prompts/s 3.7293006678555862 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 484.4740772962947 tokens/s 484.80908682122623 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3.49377375744674 prompts/s 3.5211531798996636 prompts/s 1.01
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1107.1419659972976 tokens/s 1115.8182311784044 tokens/s 1.01
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 798.9794746779702 tokens/s 804.92622718325 tokens/s 1.01
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3.7679328215644747 prompts/s 3.7651091193539057 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3862.1311421035866 tokens/s 3859.2368473377533 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 6.549583486392014 prompts/s 6.56829707726841 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 851.4458532309618 tokens/s 853.8786200448932 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 15.583125463507121 prompts/s 15.433329621058155 prompts/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 4004.8632441213304 tokens/s 3966.3657126119456 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.9147896885480673 prompts/s 0.9145165537321124 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 118.92265951124875 tokens/s 118.88715198517461 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1.0158322881571384 prompts/s 1.0166193828390906 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 132.058197460428 tokens/s 132.16051976908176 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3.7685508044012104 prompts/s 3.7643120981278515 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3862.7645745112404 tokens/s 3858.419900581048 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 7.500979459257388 prompts/s 7.4954504676739 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3848.00246259904 tokens/s 3845.166089916711 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.2563898331192543 prompts/s 0.2566596724227775 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 33.330678305503056 tokens/s 33.365757414961074 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 7.25659226026052 prompts/s 7.180857634419535 prompts/s 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3392.892277207409 tokens/s 3357.481795549198 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\": 256,\n \"output-len\": 1,\n \"num-prompts\": 1,\n \"sparsity\": \"semi_structured_sparse_w16a16\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 13.906612093571963 prompts/s 13.962956945596787 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3573.9993080479944 tokens/s 3588.4799350183744 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 0.4918983572833268 prompts/s 0.49190834701485375 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 130.0284117642746 tokens/s 130.03105244990644 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 121.98751328388316 tokens/s 121.99654944866384 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 24.123599287509897 prompts/s 24.187146791329344 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3111.944308088777 tokens/s 3120.1419360814853 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: a10b831 Previous: 93183d6 Ratio
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 7689.986569999974 ms 7772.880707999889 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 155.2369954986707 ms 156.48812977466272 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 127.68052250009987 ms 133.6518690000048 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 57.18177578822051 ms 57.264786195731034 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 56.90111482577333 ms 56.87179617812733 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 6355.970685499983 ms 6371.696383000028 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 111.16883072000064 ms 110.48925968000124 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 70.46714699993117 ms 69.49467699996603 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_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 36.17667501279266 ms 36.201560438764346 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 36.897573278567734 ms 36.90810849399123 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 13543.124993999754 ms 13774.644650499795 ms 0.98
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 198.0917725726455 ms 206.66789126134367 ms 0.96
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 157.69420799961154 ms 160.7057569999597 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 106.5214924319586 ms 109.51795528662116 ms 0.97
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 111.27711537817672 ms 113.96346420574625 ms 0.98
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 14884.52258549978 ms 15245.16342600009 ms 0.98
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 215.83828740533318 ms 216.33090558799086 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 170.05572449988904 ms 170.93354750022627 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 119.93837186669334 ms 123.57023909569018 ms 0.97
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 125.709955622744 ms 128.89090979804146 ms 0.98
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 59568.68839799995 ms 60296.60475000003 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 37536.996530118005 ms 38417.961550126 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 36862.34628149998 ms 38246.879616499995 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 97.34058150278607 ms 98.11643894172259 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 99.78903259935451 ms 100.4101346244016 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 76216.40349400013 ms 76459.87004700009 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 56605.960194843996 ms 57006.08940392401 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 69334.1744034999 ms 69956.17225149977 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 67.27457006131601 ms 66.29421221482295 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 65.15658977573332 ms 64.44521783417453 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2413.9468949997536 ms 2410.6342294999195 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 115.32430701464425 ms 116.91471860398451 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 82.14013299993894 ms 82.25686200012206 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 17.99891797558831 ms 18.08641513766197 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 16.15897576666761 ms 16.237355544863274 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 5087.44845199999 ms 5068.800384000497 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 88.80579291998099 ms 89.11502234658353 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 63.061083001230145 ms 62.95106050038157 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 36.240965857212494 ms 36.22859554011948 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 30.85187867530749 ms 30.890091700003964 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1943.3179324996672 ms 1938.3338355000888 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 79.02932675337676 ms 79.67521980001038 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_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 39.14830800022173 ms 40.12038950031638 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_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 11.069767070678836 ms 11.079684189198728 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 11.176234212828936 ms 11.16000790993628 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 6100.955769499933 ms 6115.162760999965 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 104.57615327331648 ms 104.56159974003337 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 65.58089699979064 ms 66.11591900013991 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_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 34.58406245663734 ms 34.617765827755456 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 35.07420007437409 ms 35.101717808082036 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 5270.538775500427 ms 5305.091374999392 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 181.91739862662266 ms 182.49676534935378 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 173.0050875003144 ms 173.33045750001475 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 39.47199793280612 ms 39.56192385121597 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 36.66351316804006 ms 36.86431237682397 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 3635.538525999891 ms 3649.4989170000736 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 143.14882953669135 ms 142.08588893001132 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_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 93.81988950008235 ms 93.60186949970739 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_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 24.17861995539896 ms 24.37970848197325 ms 0.99
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 23.656972808143404 ms 23.666738033050542 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 5823.356046999834 ms 5821.214592499928 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 119.88542066930434 ms 118.52209673600494 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 89.16509849950671 ms 90.16225599953032 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 53.13534165651794 ms 53.09503225674646 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 46.2948623041938 ms 46.1659372870581 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2552.110619999439 ms 2590.1457790000677 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 120.35040164531893 ms 122.22204088800088 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 83.91137899980095 ms 84.77577700023176 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_\": \"750,2.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 19.003036375024585 ms 19.27610465963713 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 17.2926273737124 ms 17.255476069253277 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 16172.425026000383 ms 16063.256238000577 ms 1.01
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 274.12945982464953 ms 285.1757610933564 ms 0.96
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 160.27451699937956 ms 163.56607699981396 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 147.62205870226447 ms 148.20962904427824 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 135.15480796223878 ms 135.77424767327517 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1921.1439794999023 ms 1921.9436020002831 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 95.49463854331407 ms 96.61242698001236 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_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 59.16418050037464 ms 61.55323299981319 ms 0.96
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 13.252081158818903 ms 13.216317727657616 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 12.528253547893332 ms 12.480939822172344 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 254997.02183650038 ms 256771.280315 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_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 240218.253697654 ms 241967.81838042397 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_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 241757.30261750004 ms 243518.75813349965 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_\": \"1500,5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 69.71335484664382 ms 68.92201654990878 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 65.96072925992397 ms 66.57899807596435 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_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 6035.140677500067 ms 6054.55432500014 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 127.43279252667244 ms 128.47732568000956 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 84.55342100000962 ms 90.02754549999281 ms 0.94
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 40.28536140570839 ms 40.41897263375693 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 40.236489911805975 ms 40.26074101871709 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 6032.537427500074 ms 6061.561180500121 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 121.9273855433418 ms 120.74932165665511 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_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 82.7403040000263 ms 75.64902700005405 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_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 39.52752941719753 ms 39.64483831073297 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 39.282266362386494 ms 39.25501033452916 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 1822.5031120000494 ms 1826.2239115001648 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 92.59179489331775 ms 93.34109191331663 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_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 57.79737500006377 ms 55.82000150025124 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_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 12.556154617036514 ms 12.62939598497101 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_\": \"300,1\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 11.89281882411575 ms 11.92246008794239 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 64642.84791799946 ms 62754.508411499046 ms 1.03
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 28202.068138578023 ms 27318.900191153323 ms 1.03
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 26515.03411000067 ms 25762.201995500618 ms 1.03
{"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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 187.5715645663519 ms 185.7308705063005 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 198.60651942622803 ms 196.39547295298112 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 2030.1320085000043 ms 2027.2287369994046 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 80.92254640001556 ms 81.79010776007392 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_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 39.536833000056504 ms 40.287139500378544 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_\": \"150,0.5\",\n \"dataset\": \"sharegpt\"\n}", "gpu_description": "NVIDIA A10G x 1", "vllm_version": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 11.594711202005476 ms 11.60009808279639 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": "0.3.0", "python_version": "3.10.12 (main, May 10 2024, 13:42:25) [GCC 9.4.0]", "torch_version": "2.3.0+cu121"} 11.670738943688308 ms 11.687424613845359 ms 1.00

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

Please sign in to comment.