refactoring: port customized kernels with public cutlass version

[yunruis]

pytest -s tests/unittest/_torch/modules/test_fused_moe.py# refactoring: port customized kernels with public cutlass version

In this PR we have open-sourced some internal Cutlass kernels. Meanwhile, to ensure stability and provide an even more optimized performance experience, we have retained the previous method of calling these kernels via static libraries, as another choice.
The purpose of this document is to introduce how to use these newly open-sourced Cutlass kernels, while also supporting the option to switch back to the previously supported internal Cutlass kernels via static libraries.

Compilation
These open-sourced cutlass kernels are low_latency_gemm, moe-gemm, fp4_gemm and allreduce_gemm. The switch between using open-sourced Cutlass kernels and static library Cutlass kernels can be made using the macro USING_OSS_CUTLASS_* achieving kernel-level control. By default, the open-source Cutlass kernels are used. For example:

python3 ./scripts/build_wheel.py --skip_building_wheel --linking_install_binary --use_ccache  --cuda_architectures "90-real;100-real" --python_bindings --install --micro_benchmarks

This will using open-sourced cutlass kernels.

If users prefer to use the internal Cutlass kernels from the static library, they can control this during compilation by setting marco USING_OSS_CUTLASS_* to OFF. For instance, if a user wants to use the static library implementation for low_latency_gemm and fused_moe_gemm, the following compilation command can be used:

python3 ./scripts/build_wheel.py --skip_building_wheel --linking_install_binary --use_ccache  --cuda_architectures "90-real;100-real" -D "USING_OSS_CUTLASS_MOE_GEMM=OFF;USING_OSS_CUTLASS_LOW_LATENCY_GEMM=OFF" --python_bindings --install --micro_benchmarks

[yunruis]

/bot run

[tensorrt-cicd]

PR_Github #8271 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #8271 [ run ] completed with state ABORTED

[yunruis]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #8280 [ run ] triggered by Bot

[yunruis]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #8282 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #8280 [ run ] completed with state ABORTED

[juney-nvidia]

Thanks for the hard working, Yunrui.
Let's merge this PR ASAP to unblock the dependency and keep refining it in the subsequent PRs.

[yunruis]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #8317 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #8282 [ run ] completed with state ABORTED
/LLM/main/L0_MergeRequest_PR pipeline #5995 completed with status: 'FAILURE'

[tensorrt-cicd]

PR_Github #8317 [ run ] completed with state ABORTED

[yunruis]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #8324 [ run ] triggered by Bot

[yunruis]

/bot kill

[tensorrt-cicd]

PR_Github #8324 [ run ] completed with state ABORTED

[yunruis]

/bot run

[tensorrt-cicd]

PR_Github #8594 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #8594 [ run ] completed with state ABORTED

[juney-nvidia]

/bot help

[github-actions]

GitHub Bot Help

/bot [-h] ['run', 'kill', 'skip', 'reuse-pipeline'] ...

Provide a user friendly way for developers to interact with a Jenkins server.

Run /bot [-h|--help] to print this help message.

See details below for each supported subcommand.

run [--disable-fail-fast --skip-test --stage-list "A10-1, xxx" --gpu-type "A30, H100_PCIe" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-[Post-Merge]-1, xxx"]

Launch build/test pipelines. All previously running jobs will be killed.

--disable-fail-fast (OPTIONAL) : Disable fail fast on build/tests/infra failures.

--skip-test (OPTIONAL) : Skip all test stages, but still run build stages, package stages and sanity check stages. Note: Does NOT update GitHub check status.

--stage-list "A10-1, xxx" (OPTIONAL) : Only run the specified test stages. Examples: "A10-1, xxx". Note: Does NOT update GitHub check status.

--gpu-type "A30, H100_PCIe" (OPTIONAL) : Only run the test stages on the specified GPU types. Examples: "A30, H100_PCIe". Note: Does NOT update GitHub check status.

--only-multi-gpu-test (OPTIONAL) : Only run the multi-GPU tests. Note: Does NOT update GitHub check status.

--disable-multi-gpu-test (OPTIONAL) : Disable the multi-GPU tests. Note: Does NOT update GitHub check status.

--add-multi-gpu-test (OPTIONAL) : Force run the multi-GPU tests. Will also run L0 pre-merge pipeline.

--post-merge (OPTIONAL) : Run the L0 post-merge pipeline instead of the ordinary L0 pre-merge pipeline.

--extra-stage "H100_PCIe-[Post-Merge]-1, xxx" (OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-[Post-Merge]-1, xxx".

kill

kill

Kill all running builds associated with pull request.

skip

skip --comment COMMENT

Skip testing for latest commit on pull request. --comment "Reason for skipping build/test" is required. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

reuse-pipeline

reuse-pipeline

Reuse a previous pipeline to validate current commit. This action will also kill all currently running builds associated with the pull request. IMPORTANT NOTE: This is dangerous since lack of user care and validation can cause top of tree to break.

[juney-nvidia]

/bot run --stage-list "DGX_H100-4_GPUs-PyTorch-Others-1"

[tensorrt-cicd]

PR_Github #8730 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #8730 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #6333 (Partly Tested) completed with status: 'SUCCESS'

[ZhanruiSunCh]

/bot skip --comment "PR_Github #8730 and PR_Github #8594 run a full pre-merge CI Pipeline with multi GPU test"

[tensorrt-cicd]

PR_Github #8768 [ skip ] triggered by Bot

[tensorrt-cicd]

PR_Github #8768 [ skip ] completed with state SUCCESS
Skipping testing for commit 758490d