[Kernel] Tuned FP8 Kernels for Ada Lovelace #6677
Conversation
|
👋 Hi! Thank you for contributing to the vLLM project. Once the PR is approved and ready to go, please make sure to run full CI as it is required to merge (or just use auto-merge). To run full CI, you can do one of these:
🚀 |
|
Kernel Benchmark Heatmap:
Pointers on how to read the heatmap: Cutlass Op naming convention: Tile Shape : 128x64x128 |
|
Thanks for the reviews @mgoin @comaniac. Even with the fix, pytorch scaled MM is better for some shapes (mostly big Ms) and your suggestion about using pytorch for big Ms is still valid. |
github-actions
bot
added
the
ready
Head branch was pushed to by a user without write access
varun-sundar-rabindranath
force-pushed
the
varun/lovelace-tune-cutlass-fp8
branch
from
68f5b29
to
85fdd37
Compare
|
/ready |
Head branch was pushed to by a user without write access
varun-sundar-rabindranath
force-pushed
the
varun/lovelace-tune-cutlass-fp8
branch
from
85fdd37
to
8091978
Compare
varun-sundar-rabindranath
force-pushed
the
varun/lovelace-tune-cutlass-fp8
branch
from
8091978
to
163e091
Compare
varun-sundar-rabindranath
changed the title
* upstream/main: (66 commits) [Bugfix] Fix PaliGemma MMP (vllm-project#6930) [TPU] Fix greedy decoding (vllm-project#6933) [Kernel] Tuned int8 kernels for Ada Lovelace (vllm-project#6848) [Kernel] Fix marlin divide-by-zero warnings (vllm-project#6904) [ci] GHA workflow to remove ready label upon "/notready" comment (vllm-project#6921) [Kernel] Remove unused variables in awq/gemm_kernels.cu (vllm-project#6908) [Frontend] New `allowed_token_ids` decoding request parameter (vllm-project#6753) [Bugfix] Allow vllm to still work if triton is not installed. (vllm-project#6786) [TPU] Support tensor parallelism in async llm engine (vllm-project#6891) [Kernel] Fix deprecation function warnings squeezellm quant_cuda_kernel (vllm-project#6901) [Core] Reduce unnecessary compute when logprobs=None (vllm-project#6532) [Kernel] Tuned FP8 Kernels for Ada Lovelace (vllm-project#6677) [Model] Initialize support for InternVL2 series models (vllm-project#6514) [Misc] Pass cutlass_fp8_supported correctly in fbgemm_fp8 (vllm-project#6871) Add Nemotron to PP_SUPPORTED_MODELS (vllm-project#6863) [Kernel] Increase precision of GPTQ/AWQ Marlin kernel (vllm-project#6795) [TPU] Reduce compilation time & Upgrade PyTorch XLA version (vllm-project#6856) [Docs] Add RunLLM chat widget (vllm-project#6857) [Model] Initial support for BLIP-2 (vllm-project#5920) [CI/Build][Doc] Update CI and Doc for VLM example changes (vllm-project#6860) ...
Co-authored-by: Varun Sundar Rabindranath <varun@neuralmagic.com>
Good performance speedups for M <= 16
llama2-7b:llama3-8b:llama2-13b:llama2-70b:Numbers:
Benchmark Serving:Machine : L40S x 1
Command :
python3 -m vllm.entrypoints.openai.api_server --model neuralmagic/Meta-Llama-3-8B-Instruct-FP8Cutlass Kernels:
Pytorch Kernels:
Gemm benchmarkspython3 benchmarks/cutlass_benchmarks/w8a8_benchmarks.py --dtype fp8 model_bench --batch-sizes {1,16,32,64,128,256,512}<style type="text/css"></style>
PR Checklist (Click to Expand)
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.
PR Title and Classification
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:
[Bugfix]for bug fixes.[CI/Build]for build or continuous integration improvements.[Doc]for documentation fixes and improvements.[Model]for adding a new model or improving an existing model. Model name should appear in the title.[Frontend]For changes on the vLLM frontend (e.g., OpenAI API server,LLMclass, etc.)[Kernel]for changes affecting CUDA kernels or other compute kernels.[Core]for changes in the core vLLM logic (e.g.,LLMEngine,AsyncLLMEngine,Scheduler, etc.)[Hardware][Vendor]for hardware-specific changes. Vendor name should appear in the prefix (e.g.,[Hardware][AMD]).[Misc]for PRs that do not fit the above categories. Please use this sparingly.Note: If the PR spans more than one category, please include all relevant prefixes.
Code Quality
The PR need to meet the following code quality standards:
format.shto format your code.docs/source/if the PR modifies the user-facing behaviors of vLLM. It helps vLLM user understand and utilize the new features or changes.Notes for Large Changes
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
rfc-requiredand might not go through the PR.What to Expect for the Reviews
The goal of the vLLM team is to be a transparent reviewing machine. 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:
action-requiredlabel on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.Thank You
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!