[TRTLLM-4958][feat] improve multimodal workflow with Vision + LLM inflight batching

[symphonylyh]

Description

NOTE: this PR provides a problem statement and proof-of-concept solutions to the problem. We plan to address it more formally in follow-up PRs in #4799 and #5522, please take a look at them! this PR will be closed after the refactoring is completely done.

Existing workflow:

• vision forward on main process, LLM forwrad on worker process. multimodal embedding is transferred during IPC.
• many IPC D2H & H2D overhead due to the queue structure being used. Tensor pickle serialization & deserialization is involved
• vision forward is per-request run, so only allows BS=1 for vision
  
  Above is the nsys for processing 3 images

Improved workflow:

• vision & LLM in the same forward, on worker process. raw image tensor is transferred during IPC.
• allows BS > 1 for vision. although enforces vision batch to immediately finish the context/prefill phase, cannot be prefilled later.
• needs to modify attn_medata struct inside forward. the overhead from the prepare_multimodal_ifb() call is negligible from nsys profile. but caveat is this is not yet compatible with cuda graph mode.
• the IFB token capacity estimate may be incorrect during request scheduling, because vision context length is unknown at scheduling time
  

Notes:

• use TLLM_MULTIMODAL_DISAGGREGATED to test old (=1) & new (=0, by default) for now
• use llava-next as the experimental example
• --user_cuda_graph doesn't work with this IFB mode because attn_metadata.prepare() needs to be called inside forward()

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.

[symphonylyh]

/bot run

[tensorrt-cicd]

PR_Github #6509 [ run ] triggered by Bot

[tensorrt-cicd]

PR_Github #6509 [ run ] completed with state FAILURE
/LLM/main/L0_MergeRequest_PR pipeline #4764 completed with status: 'FAILURE'

[MartinMarciniszyn]

@symphonylyh , is this PR still relevant, or should we close it?