This repository has been archived by the owner on Oct 11, 2024. It is now read-only.
Upstream sync 2024 05 05 #224
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…t#4376) Co-authored-by: WoosukKwon <[email protected]>
Co-authored-by: Danny Guinther <[email protected]>
Co-authored-by: Simon Mo <[email protected]>
…les (vllm-project#4343) Co-authored-by: Woosuk Kwon <[email protected]>
Co-authored-by: Antoni Baum <[email protected]>
) Co-authored-by: DefTruth <[email protected]>
Signed-off-by: Prashant Gupta <[email protected]> Co-authored-by: Travis Johnson <[email protected]>
Co-authored-by: Robert Shaw <[email protected]> Co-authored-by: Robert Shaw <[email protected]>
…project#3922) Co-authored-by: alexm <[email protected]> Co-authored-by: mgoin <[email protected]>
FILL IN THE PR DESCRIPTION HERE
FIX #xxxx (*link existing issues this PR will resolve*)
**BEFORE SUBMITTING, PLEASE READ THE CHECKLIST BELOW AND FILL IN THE
DESCRIPTION ABOVE**
---
<details>
<!-- inside this <details> section, markdown rendering does not work, so
we use raw html here. -->
<summary><b> PR Checklist (Click to Expand) </b></summary>
<p>Thank you for your contribution to vLLM! Before submitting the pull
request, please ensure the PR meets the following criteria. This helps
vLLM maintain the code quality and improve the efficiency of the review
process.</p>
<h3>PR Title and Classification</h3>
<p>Only specific types of PRs will be reviewed. The PR title is prefixed
appropriately to indicate the type of change. Please use one of the
following:</p>
<ul>
<li><code>[Bugfix]</code> for bug fixes.</li>
<li><code>[CI/Build]</code> for build or continuous integration
improvements.</li>
<li><code>[Doc]</code> for documentation fixes and improvements.</li>
<li><code>[Model]</code> for adding a new model or improving an existing
model. Model name should appear in the title.</li>
<li><code>[Frontend]</code> For changes on the vLLM frontend (e.g.,
OpenAI API server, <code>LLM</code> class, etc.) </li>
<li><code>[Kernel]</code> for changes affecting CUDA kernels or other
compute kernels.</li>
<li><code>[Core]</code> for changes in the core vLLM logic (e.g.,
<code>LLMEngine</code>, <code>AsyncLLMEngine</code>,
<code>Scheduler</code>, etc.)</li>
<li><code>[Hardware][Vendor]</code> for hardware-specific changes.
Vendor name should appear in the prefix (e.g.,
<code>[Hardware][AMD]</code>).</li>
<li><code>[Misc]</code> for PRs that do not fit the above categories.
Please use this sparingly.</li>
</ul>
<p><strong>Note:</strong> If the PR spans more than one category, please
include all relevant prefixes.</p>
<h3>Code Quality</h3>
<p>The PR need to meet the following code quality standards:</p>
<ul>
<li>We adhere to <a
href="https://google.github.io/styleguide/pyguide.html">Google Python
style guide</a> and <a
href="https://google.github.io/styleguide/cppguide.html">Google C++
style guide</a>.</li>
<li>Pass all linter checks. Please use <a
href="https://github.com/vllm-project/vllm/blob/main/format.sh"><code>format.sh</code></a>
to format your code.</li>
<li>The code need to be well-documented to ensure future contributors
can easily understand the code.</li>
<li>Include sufficient tests to ensure the project to stay correct and
robust. This includes both unit tests and integration tests.</li>
<li>Please add documentation to <code>docs/source/</code> if the PR
modifies the user-facing behaviors of vLLM. It helps vLLM user
understand and utilize the new features or changes.</li>
</ul>
<h3>Notes for Large Changes</h3>
<p>Please keep the changes as concise as possible. For major
architectural changes (>500 LOC excluding kernel/data/config/test), we
would expect a GitHub issue (RFC) discussing the technical design and
justification. Otherwise, we will tag it with <code>rfc-required</code>
and might not go through the PR.</p>
<h3>What to Expect for the Reviews</h3>
<p>The goal of the vLLM team is to be a <i>transparent reviewing
machine</i>. We would like to make the review process transparent and
efficient and make sure no contributor feel confused or frustrated.
However, the vLLM team is small, so we need to prioritize some PRs over
others. Here is what you can expect from the review process: </p>
<ul>
<li> After the PR is submitted, the PR will be assigned to a reviewer.
Every reviewer will pick up the PRs based on their expertise and
availability.</li>
<li> After the PR is assigned, the reviewer will provide status update
every 2-3 days. If the PR is not reviewed within 7 days, please feel
free to ping the reviewer or the vLLM team.</li>
<li> After the review, the reviewer will put an <code>
action-required</code> label on the PR if there are changes required.
The contributor should address the comments and ping the reviewer to
re-review the PR.</li>
<li> Please respond to all comments within a reasonable time frame. If a
comment isn't clear or you disagree with a suggestion, feel free to ask
for clarification or discuss the suggestion.
</li>
</ul>
<h3>Thank You</h3>
<p> Finally, thank you for taking the time to read these guidelines and
for your interest in contributing to vLLM. Your contributions make vLLM
a great tool for everyone! </p>
</details>
mgoin
approved these changes
May 13, 2024
LucasWilkinson
added a commit
that referenced
this pull request
Jun 26, 2024
…ofiler not picking up kernels within cudagraphs (#332) This PR updates `examples/offline_profile.py` to use the new `add_requests` interface. In addition, this PR fixes issues with profiler not picking up kernels run from within a cudagraph (i.e. when profiling with `--allow-cuda-graphs`, there's is too main issues: 1) Changes from the initial profiler PR (#124) were wiped out by #224, namely the changes in `model_runner.py` converting `CUDAGraphRunner` to a `nn.Module` allowing the profiler to pick it up 2) Many kernels within the graph had the same correlation id so we were always picking the first of potentially many kernels to display, using name in addition to correlation Id sees to resolve this issue but is potentially fragile Before the PR: ``` ================================================================================ = Decode Summary Table (prompt_len=1, batch_size=1) ================================================================================ name | cuda_time_us | pct_cuda_... | invocations ================================================================================================================================ LogitsProcessor | 350.00 | 57.76 | 1.00 |- void at::native::(anonymous namespace)::indexSelectSmallIndex<c10::BFloat1... | 3.00 | 0.50 | 1.00 |- sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize64x128x64_warpgroupsize1x... | 347.00 | 57.26 | 1.00 Sampler | 256.00 | 42.24 | 1.00 |- Memcpy HtoD (Pinned -> Device) | 18.00 | 2.97 | 9.00 |- void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_no... | 4.00 | 0.66 | 1.00 |- void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kerne... | 5.00 | 0.83 | 2.00 |- void at::native::(anonymous namespace)::cunn_SoftMaxForward<4, float, floa... | 54.00 | 8.91 | 1.00 |- void at::native::(anonymous namespace)::cunn_SoftMaxForward<4, float, floa... | 43.00 | 7.10 | 1.00 |- void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kerne... | 4.00 | 0.66 | 2.00 |- void at::native::index_elementwise_kernel<128, 4, at::native::gpu_index_ke... | 13.00 | 2.15 | 4.00 |- void at::native::reduce_kernel<512, 1, at::native::ReduceOp<float, at::nat... | 29.00 | 4.79 | 1.00 |- Memcpy DtoH (Device -> Pageable) | 10.00 | 1.65 | 5.00 |- void (anonymous namespace)::elementwise_kernel_with_index<int, at::native:... | 1.00 | 0.17 | 1.00 |- void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_no... | 3.00 | 0.50 | 1.00 |- void at::native::reduce_kernel<512, 1, at::native::ReduceOp<long, at::nati... | 15.00 | 2.48 | 1.00 |- void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctorO... | 1.00 | 0.17 | 1.00 |- void at::native::mbtopk::fill<unsigned int, unsigned int>(unsigned int*, u... | 1.00 | 0.17 | 1.00 |- void at::native::mbtopk::radixFindKthValues<float, unsigned int, unsigned ... | 36.00 | 5.94 | 4.00 |- void at::native::mbtopk::computeBlockwiseWithinKCounts<unsigned int>(unsig... | 8.00 | 1.32 | 4.00 |- void at::native::mbtopk::computeBlockwiseKthCounts<unsigned int>(unsigned ... | 1.00 | 0.17 | 1.00 |- void at_cuda_detail::cub::DeviceScanByKeyInitKernel<at_cuda_detail::cub::R... | 2.00 | 0.33 | 2.00 |- void at_cuda_detail::cub::DeviceScanByKeyKernel<at_cuda_detail::cub::Devic... | 4.00 | 0.66 | 2.00 |- void at::native::mbtopk::gatherTopK<float, unsigned int, 1>(at::cuda::deta... | 4.00 | 0.66 | 1.00 ``` After PR ``` name | cuda_time_us | pct_cuda_... | invocations ================================================================================================================================ CUDAGraphRunner | 4238.00 | 84.41 | 1.00 |- Memcpy DtoD (Device -> Device) | 5.00 | 0.10 | 5.00 |- void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<... | 2.00 | 0.04 | 2.00 |- void at::native::(anonymous namespace)::indexSelectSmallIndex<c10::BFloat1... | 2.00 | 0.04 | 1.00 |- void vllm::rms_norm_kernel<c10::BFloat16>(c10::BFloat16*, c10::BFloat16 co... | 4.00 | 0.08 | 1.00 |- void vllm::scaled_fp8_quant_kernel<c10::BFloat16>(c10::Float8_e4m3fn*, c10... | 256.00 | 5.10 | 128.00 |- sm90_xmma_gemm_e4m3bf16_e4m3f32_f32_tn_n_tilesize64x64x128_warpgroupsize1x... | 1440.00 | 28.68 | 96.00 |- void vllm::rotary_embedding_kernel<c10::BFloat16, true>(long const*, c10::... | 96.00 | 1.91 | 32.00 |- void vllm::reshape_and_cache_flash_kernel<c10::BFloat16>(c10::BFloat16 con... | 64.00 | 1.27 | 32.00 |- void flash_fwd_splitkv_kernel<Flash_fwd_kernel_traits<128, 64, 128, 4, fal... | 160.00 | 3.19 | 32.00 |- void flash_fwd_splitkv_combine_kernel<Flash_fwd_kernel_traits<128, 64, 128... | 160.00 | 3.19 | 32.00 |- memcpy32_post | 33.00 | 0.66 | 33.00 |- std::enable_if<(((8)>(0)))&&vllm::_typeConvert<c10::BFloat16>::exists, voi... | 128.00 | 2.55 | 64.00 |- sm90_xmma_gemm_e4m3bf16_e4m3f32_f32_tn_n_tilesize64x128x128_warpgroupsize1... | 1664.00 | 33.14 | 32.00 |- void vllm::act_and_mul_kernel<c10::BFloat16, &(c10::BFloat16 vllm::silu_ke... | 224.00 | 4.46 | 32.00 LogitsProcessor | 351.00 | 6.99 | 1.00 |- void at::native::(anonymous namespace)::indexSelectSmallIndex<c10::BFloat1... | 3.00 | 0.06 | 1.00 |- sm90_xmma_gemm_bf16bf16_bf16f32_f32_tn_n_tilesize64x128x64_warpgroupsize1x... | 348.00 | 6.93 | 1.00 Sampler | 432.00 | 8.60 | 1.00 |- Memcpy HtoD (Pinned -> Device) | 18.00 | 0.36 | 9.00 |- void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_no... | 4.00 | 0.08 | 1.00 |- at::native::(anonymous namespace)::fill_reverse_indices_kernel(long*, int,... | 2.00 | 0.04 | 1.00 |- void at_cuda_detail::cub::DeviceRadixSortUpsweepKernel<at_cuda_detail::cub... | 12.00 | 0.24 | 3.00 |- void at_cuda_detail::cub::RadixSortScanBinsKernel<at_cuda_detail::cub::Dev... | 7.00 | 0.14 | 3.00 |- void at_cuda_detail::cub::DeviceRadixSortDownsweepKernel<at_cuda_detail::c... | 46.00 | 0.92 | 3.00 |- Memcpy DtoD (Device -> Device) | 1.00 | 0.02 | 1.00 |- void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kerne... | 5.00 | 0.10 | 3.00 |- void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctorO... | 1.00 | 0.02 | 1.00 |- void at::native::_scatter_gather_elementwise_kernel<128, 4, at::native::_c... | 6.00 | 0.12 | 2.00 |- void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_no... | 6.00 | 0.12 | 2.00 |- void at::native::vectorized_elementwise_kernel<4, at::native::(anonymous n... | 2.00 | 0.04 | 2.00 |- void at::native::(anonymous namespace)::cunn_SoftMaxForward<8, c10::BFloat... | 39.00 | 0.78 | 1.00 |- void at_cuda_detail::cub::DeviceScanInitKernel<at_cuda_detail::cub::ScanTi... | 1.00 | 0.02 | 1.00 |- void at_cuda_detail::cub::DeviceScanKernel<at_cuda_detail::cub::DeviceScan... | 6.00 | 0.12 | 1.00 |- void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctorO... | 1.00 | 0.02 | 1.00 |- void at::native::unrolled_elementwise_kernel<at::native::FillFunctor<bool>... | 1.00 | 0.02 | 1.00 |- void (anonymous namespace)::elementwise_kernel_with_index<int, at::native:... | 3.00 | 0.06 | 2.00 |- void at::native::_scatter_gather_elementwise_kernel<128, 4, at::native::_c... | 5.00 | 0.10 | 1.00 |- void at::native::unrolled_elementwise_kernel<at::native::direct_copy_kerne... | 5.00 | 0.10 | 2.00 |- void at::native::(anonymous namespace)::cunn_SoftMaxForward<4, float, floa... | 86.00 | 1.71 | 1.00 |- void at::native::(anonymous namespace)::cunn_SoftMaxForward<4, float, floa... | 43.00 | 0.86 | 1.00 |- void at::native::index_elementwise_kernel<128, 4, at::native::gpu_index_ke... | 14.00 | 0.28 | 4.00 |- void at::native::(anonymous namespace)::distribution_elementwise_grid_stri... | 3.00 | 0.06 | 1.00 |- void at::native::vectorized_elementwise_kernel<4, at::native::BinaryFuncto... | 2.00 | 0.04 | 1.00 |- void at::native::reduce_kernel<512, 1, at::native::ReduceOp<float, at::nat... | 28.00 | 0.56 | 1.00 |- Memcpy DtoH (Device -> Pageable) | 10.00 | 0.20 | 5.00 |- void at::native::elementwise_kernel<128, 4, at::native::gpu_kernel_impl_no... | 3.00 | 0.06 | 1.00 |- void at::native::reduce_kernel<512, 1, at::native::ReduceOp<long, at::nati... | 15.00 | 0.30 | 1.00 |- void at::native::vectorized_elementwise_kernel<4, at::native::CUDAFunctorO... | 1.00 | 0.02 | 1.00 |- void at::native::mbtopk::fill<unsigned int, unsigned int>(unsigned int*, u... | 1.00 | 0.02 | 1.00 |- void at::native::mbtopk::radixFindKthValues<float, unsigned int, unsigned ... | 36.00 | 0.72 | 4.00 |- void at::native::mbtopk::computeBlockwiseWithinKCounts<unsigned int>(unsig... | 8.00 | 0.16 | 4.00 |- void at::native::mbtopk::computeBlockwiseKthCounts<unsigned int>(unsigned ... | 1.00 | 0.02 | 1.00 |- void at_cuda_detail::cub::DeviceScanByKeyInitKernel<at_cuda_detail::cub::R... | 2.00 | 0.04 | 2.00 |- void at_cuda_detail::cub::DeviceScanByKeyKernel<at_cuda_detail::cub::Devic... | 4.00 | 0.08 | 2.00 |- void at::native::mbtopk::gatherTopK<float, unsigned int, 1>(at::cuda::deta... | 4.00 | 0.08 | 1.00 ``` --------- Co-authored-by: Lucas Wilkinson <[email protected]>
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Upstream sync 2024 05 25 (#224) - release v0.4.2
SUMMARY:
Merge commits from vllm-project@b6dcb4d to vllm-project@c7f2cf2
Note that vllm-project@b6dcb4d is NOT included in this merge.