Mtp optimizations round1

[ameynaik-hub]

Deepseek MTP optimization.

Optimize mtp head for deepseek

1. eh_proj gemm sharding
2. rmsnorm cuda stream parallelization.
3. lm_head top1 using local argmax --> AG --> argmax
4. piecewise torch.compile to fuse / opt multiple native kernels.

Test Coverage

BS1 deepseek perf improves from 411-->431 TPS on B200x8 TP8EP1.

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".

For guidance on mapping tests to stage names, see docs/source/reference/ci-overview.md.

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.

Summary by CodeRabbit

• New Features

  • Improved performance and efficiency for token state extraction and normalization in DeepseekV3 models.
  • Enhanced support for distributed and parallel token selection in speculative decoding with tensor parallelism.
  • Added new helper functions for optimized token sampling and draft token management.

• Refactor

  • Modularized and compiled key tensor operations for better clarity and speed.
  • Updated constructors and initialization logic to consistently use model configuration context across relevant components.

• Chores

  • Adjusted internal interfaces to support expanded configuration and parallelism features.

[ameynaik-hub]

/bot run

[Kefeng-Duan]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #11430 [ run ] triggered by Bot

[Kefeng-Duan]

@lfr-0531 @yweng0828 for viz

[tensorrt-cicd]

PR_Github #11430 [ run ] completed with state FAILURE
/LLM/main/L0_MergeRequest_PR pipeline #8454 completed with status: 'FAILURE'

[ameynaik-hub]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #11459 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #11459 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #8478 completed with status: 'FAILURE'

[ameynaik-hub]

/bot run --disable-fail-fast

[ameynaik-hub]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #11905 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #11905 [ run ] completed with state FAILURE
/LLM/main/L0_MergeRequest_PR pipeline #8824 completed with status: 'FAILURE'

[ameynaik-hub]

/bot run --disable-fail-fast

[Kefeng-Duan]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #12670 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #12670 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #9421 completed with status: 'FAILURE'

[hello-11]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #12788 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #12788 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #9526 completed with status: 'FAILURE'

[Kefeng-Duan]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #12824 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #12824 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #9558 completed with status: 'FAILURE'

[Kefeng-Duan]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #12860 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #12860 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #9584 completed with status: 'SUCCESS'
Pipeline passed with automatic retried tests. Check the rerun report for details.

[QiJune]

/bot skip --comment "reuse previous passed CI"

[tensorrt-cicd]

PR_Github #12959 [ skip ] triggered by Bot

[tensorrt-cicd]

PR_Github #12959 [ skip ] completed with state ABORTED

[Kefeng-Duan]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #12991 [ run ] triggered by Bot

[ameynaik-hub]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #13014 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #13014 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #9718 completed with status: 'SUCCESS'