[Kernel] Turn off CUTLASS scaled_mm for Ada Lovelace

[tlrmchlsmth]

We haven't tuned the cutlass kernels for SM89 and performance is very bad.

The following benchmarks were run on an L40 with

    python benchmarks/benchmark_serving.py \
    --backend openai \
    --model neuralmagic/Meta-Llama-3-8B-Instruct-FP8 \
    --dataset-path ShareGPT_V3_unfiltered_cleaned_split.json \
    --request-rate 1 \
    --num-prompts 200 \
    --port 8000

Using the CUTLASS kernels:

============ Serving Benchmark Result ============
Successful requests:                     200
Benchmark duration (s):                  211.44
Total input tokens:                      42659
Total generated tokens:                  41049
Request throughput (req/s):              0.95
Input token throughput (tok/s):          201.76
Output token throughput (tok/s):         194.14
---------------Time to First Token----------------
Mean TTFT (ms):                          71.19
Median TTFT (ms):                        63.17
P99 TTFT (ms):                           159.21
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          35.87
Median TPOT (ms):                        35.42
P99 TPOT (ms):                           46.35
---------------Inter-token Latency----------------
Mean ITL (ms):                           35.67
Median ITL (ms):                         33.55
P99 ITL (ms):                            102.87
==================================================

Using torch.scaled_mm:

============ Serving Benchmark Result ============
Successful requests:                     200
Benchmark duration (s):                  202.07
Total input tokens:                      42659
Total generated tokens:                  40851
Request throughput (req/s):              0.99
Input token throughput (tok/s):          211.11
Output token throughput (tok/s):         202.16
---------------Time to First Token----------------
Mean TTFT (ms):                          30.50
Median TTFT (ms):                        29.03
P99 TTFT (ms):                           60.73
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          16.58
Median TPOT (ms):                        16.46
P99 TPOT (ms):                           19.92
---------------Inter-token Latency----------------
Mean ITL (ms):                           16.52
Median ITL (ms):                         16.09
P99 ITL (ms):                            33.85
==================================================

---

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, LLM class, 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:

• We adhere to Google Python style guide and Google C++ style guide.
• Pass all linter checks. Please use format.sh to format your code.
• The code need to be well-documented to ensure future contributors can easily understand the code.
• Include sufficient tests to ensure the project to stay correct and robust. This includes both unit tests and integration tests.
• Please add documentation to 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-required and 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:

• 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.
• 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.
• After the review, the reviewer will put an action-required label on the PR if there are changes required. The contributor should address the comments and ping the reviewer to re-review the PR.
• 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.

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!

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
Just a reminder: PRs would not trigger full CI run by default. Instead, it would only trigger fastcheck CI to run, which consists only a small and essential subset of tests to quickly catch small errors.

Full CI run is still required to merge this PR so please make sure that you run full CI before merging or if you need more test signals.

To run full CI, you can do one of these:

• Add ready label to the PR
• Comment /ready on the PR
• Enable auto-merge.

🚀

[tlrmchlsmth]

This should fix #6240

In a future PR we'll tune the CUTLASS SM89 FP8 kernels. We will need to do this anyway for the int8 case as well.

[tlrmchlsmth]

/ready

[robertgshaw2-neuralmagic]

@tlrmchlsmth FYI - the lm-eval regression tests failed because the score for fp8 llama changed a bit for fp8 with this change. I just tweaked the values in the config.

Note: we are only running 250 samples for this model. I will re-run with 1000 samples so its a bit more stable in a separate PR

[tlrmchlsmth]

@robertgshaw2-neuralmagic thanks. Looks like we're failing on the GPTQ marlin tests as well. Are those flakey?

[robertgshaw2-neuralmagic]

@robertgshaw2-neuralmagic thanks. Looks like we're failing on the GPTQ marlin tests as well. Are those flakey?

yes due to nondeteminism. will rerun