feat: chunked prefill for MLA (Blackwell)

[jmydurant]

[TRTLLM-3602][feat]Draft: chunked prefill for MLA (Blackwell)

Description

This PR is to support chunked context for MLA. In order to save GPU memory, we need make KV cache into piece of small chunk for each round and merge the attention output with lse.

This is a draft version and under construction

Test Coverage

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.

[poweiw]

Hello @jmydurant! Please ignore if I'm wrong but could you finish the NVIDIA github onboarding process? We're not seeing you in the NVIDIA members and this will cause false failures for community identifiers.

[kaiyux]

/bot run

[tensorrt-cicd]

PR_Github #9079 [ run ] triggered by Bot

[jmydurant]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #9324 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #9327 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #9324 [ run ] completed with state ABORTED

[jmydurant]

/bot kill

[jmydurant]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #9793 [ run ] triggered by Bot

[kaiyux]

@NVIDIA/trt-llm-torch-devs can you help review this PR as well? Thanks.

[jmydurant]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #9810 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #9793 [ run ] completed with state ABORTED

[tensorrt-cicd]

PR_Github #9810 [ run ] completed with state FAILURE
/LLM/main/L0_MergeRequest_PR pipeline #7237 completed with status: 'FAILURE'

[jmydurant]

/bot run --disable-fail-fast

[tensorrt-cicd]

PR_Github #9872 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #9872 [ run ] completed with state SUCCESS
/LLM/main/L0_MergeRequest_PR pipeline #7284 completed with status: 'SUCCESS'

[Shang-Pin]

Very excited of chunked prefill support for mla. I want to report a bug I saw when I testing this branch. The first request will return successfully, but the subsequent requests will never terminate and continue running forever.

Edit: I think it might be caused by enabling fp8 kv cache, it works when disabled.

[kaiyux]

Very excited of chunked prefill support for mla. I want to report a bug I saw when I testing this branch. The first request will return successfully, but the subsequent requests will never terminate and continue running forever.

Edit: I think it might be caused by enabling fp8 kv cache, it works when disabled.

@Shang-Pin Thanks a lot for your attention and help reporting the issue! The fp8 kv cache with chunked MLA support on Blackwell is going to be added in #5475.