[SGLANG] add sglang block fp8 gemm kernels into benchmark

[airMeng]

New contributor declaration

• I am not making a trivial change, such as fixing a typo in a comment.

• I have written a PR description following these
  We are working on SGLang upstream, and encounter a lot of issues on triton side. Following [SGLANG][Triton 3.4] Functional enabling and performance benchmarking #3622, we will add SGLang kernels into Triton benchmark suite to monitor its functionality and performance.

• I have run pre-commit run --from-ref origin/main --to-ref HEAD.

• Select one of the following.

  • I have added tests.
    • /test for lit tests
    • /unittest for C++ tests
    • /python/test for end-to-end tests
  • This PR does not need a test because this is improvements of benchmark targets.

• Select one of the following.

  • I have not added any lit tests.
  • The lit tests I have added follow these best practices,
    including the "tests should be minimal" section. (Usually running Python code
    and using the instructions it generates is not minimal.)

[airMeng]

@vlad-penkin we are new to the project, could you let me know whether the PR is acceptable? Of course we should wait for the related functionality ready then merge this?

[leonling-ll]

There are need some changes to make it functionally work.

To integrate this kernel to Triton regular benchmark, you can refer to triton-benchmarks.yml#L136

You can run python3 -m pre_commit run --show-diff-on-failure --color=always --all-files --verbose to check and pass the code style check.

[leonling-ll]

@vlad-penkin Can you help confirm is there any licence concern on porting public kernels to benchmark?

[vlad-penkin]

Let's not vendor in sglang. There are multiple technical reasons why. We should follow the same integration approach and principles as for

• PyTorch. We pin commit id and have patch mechanism to apply open PR's in PT repo : https://github.com/intel/intel-xpu-backend-for-triton/blob/main/scripts/install-pytorch.sh
• Liger-Kernels: https://github.com/intel/intel-xpu-backend-for-triton/blob/main/.github/workflows/third-party-benchmarks.yml

[airMeng]

@vlad-penkin I agree to put the benchmark under the same principles so I will close this PR. When would the whole SGLang integrated?

[mingfeima]

is this to say we need to install sglang and test the benchmark from sglang, even in this repo ?

[whitneywhtsang]

@airMeng can we close this PR and continue in #3796?

[whitneywhtsang]

Let's reopen if needed.