[TRTLLM-10064][feat] Support multi-stream for mamba prefill/decode #10291
Conversation
Wanli-Jiang
force-pushed
the
user/williamj/support-multistream-mamba-pd
branch
2 times, most recently
from
ec783d5 to
3e16819
Compare
📝 WalkthroughWalkthroughAdds multi-stream parallel execution support to Mamba2Mixer by introducing an aux_stream_dict parameter. NemotronHModel now dynamically creates auxiliary streams for Attention, MoeShared, MoeChunkingOverlap, and MoeBalancer types and passes them to Mamba2Mixer, which uses separate helper methods for prefill and decode computations that can run in parallel. Changes
Sequence Diagram(s)sequenceDiagram
participant NemotronH as NemotronHModel
participant Mamba2 as Mamba2Mixer
participant Stream as torch.cuda.Stream
participant Parallel as maybe_execute_in_parallel
NemotronH->>NemotronH: Create aux_stream_dict<br/>(Attention, MoeShared,<br/>MoeChunkingOverlap,<br/>MoeBalancer)
rect rgba(100, 150, 200, 0.3)
Note over NemotronH,Stream: Stream Initialization
NemotronH->>Stream: Create Stream for each<br/>AuxStreamType
NemotronH->>Mamba2: Pass aux_stream_dict<br/>+ config
end
rect rgba(150, 200, 100, 0.3)
Note over Mamba2,Parallel: Forward Pass (Multi-Stream Mode)
Mamba2->>Mamba2: Separate prefill/<br/>decode inputs
Mamba2->>Parallel: Launch parallel execution
Parallel->>Mamba2: _prefill_forward<br/>(on aux stream)
Parallel->>Mamba2: _decode_forward<br/>(on aux stream)
Mamba2->>Mamba2: Synchronize streams<br/>& concatenate results
Mamba2->>Mamba2: Copy decoded ssm_states<br/>back to shared tensor
end
Mamba2-->>NemotronH: Return concatenated<br/>output
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes Pre-merge checks and finishing touches❌ Failed checks (2 warnings)
✅ Passed checks (1 passed)
✨ Finishing touches
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 0
🧹 Nitpick comments (2)
tensorrt_llm/_torch/models/modeling_nemotron_h.py (1)
1-1: Update copyright year to include 2025.As per coding guidelines, all TensorRT-LLM Open Source Software code should contain an NVIDIA copyright header that includes the year of its latest meaningful modification. This file has been modified in 2025.
Proposed fix
-# SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-FileCopyrightText: Copyright (c) 2022-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.tensorrt_llm/_torch/modules/mamba/mamba2_mixer.py (1)
1-1: Update copyright year to include 2025.As per coding guidelines, the copyright header should include the year of its latest meaningful modification.
Proposed fix
-# SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved. +# SPDX-FileCopyrightText: Copyright (c) 2022-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (2)
tensorrt_llm/_torch/models/modeling_nemotron_h.pytensorrt_llm/_torch/modules/mamba/mamba2_mixer.py
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Code developed for TensorRT-LLM should conform to Python 3.8+
Indent Python code with 4 spaces. Do not use tabs
Always maintain the namespace when importing in Python, even if only one class or function from a module is used
Python files should use snake_case naming:some_file.py
Python classes should use PascalCase naming:class SomeClass
Python functions and methods should use snake_case naming:def my_awesome_function():
Python local variables should use snake_case naming:my_variable = ...
Python variable names that start with a number should be prefixed with 'k':k_99th_percentile = ...
Python global variables should use upper snake_case with prefix 'G':G_MY_GLOBAL = ...
Python constants should use upper snake_case naming:MY_CONSTANT = ...
Avoid shadowing variables declared in an outer scope in Python
Initialize all externally visible members of a Python class in the constructor
For Python interfaces that may be used outside a file, prefer docstrings over comments
Python comments should be reserved for code within a function, or interfaces that are local to a file
Use Google style docstrings in Python for classes and functions, which can be parsed by Sphinx
Python attributes and variables can be documented inline with type and description
Avoid using reflection in Python when functionality can be easily achieved without reflection
When using try-except blocks in Python, limit the except to the smallest set of errors possible
When using try-except blocks in Python to handle multiple possible variable types (duck-typing), keep the body of the try as small as possible, using the else block for logic
Files:
tensorrt_llm/_torch/modules/mamba/mamba2_mixer.pytensorrt_llm/_torch/models/modeling_nemotron_h.py
**/*.{cpp,h,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
All TensorRT-LLM Open Source Software code should contain an NVIDIA copyright header that includes the year of its latest meaningful modification
Files:
tensorrt_llm/_torch/modules/mamba/mamba2_mixer.pytensorrt_llm/_torch/models/modeling_nemotron_h.py
🧬 Code graph analysis (2)
tensorrt_llm/_torch/modules/mamba/mamba2_mixer.py (3)
tensorrt_llm/_torch/compilation/multi_stream/auto_multi_stream.py (1)
Stream(89-97)tensorrt_llm/_torch/modules/mamba/mamba2_metadata.py (1)
Mamba2Metadata(88-155)tensorrt_llm/_torch/modules/mamba/ssd_combined.py (1)
mamba_chunk_scan_combined(183-252)
tensorrt_llm/_torch/models/modeling_nemotron_h.py (2)
tensorrt_llm/_torch/models/modeling_utils.py (1)
config(526-527)tensorrt_llm/_torch/compilation/multi_stream/auto_multi_stream.py (1)
Stream(89-97)
⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (1)
- GitHub Check: Pre-commit Check
🔇 Additional comments (8)
tensorrt_llm/_torch/models/modeling_nemotron_h.py (2)
349-359: LGTM!The
aux_stream_dictcreation is clean and well-documented. The comment at line 350-351 clarifies the purpose of the Attention stream for Mamba2Mixer prefill/decode operations.
302-314: LGTM!The
Mamba2Mixerinstantiation correctly passes theconfigandaux_stream_dictparameters, enabling multi-stream parallel execution for prefill/decode operations.tensorrt_llm/_torch/modules/mamba/mamba2_mixer.py (6)
16-28: LGTM!The new imports are correctly added and all are used in the implementation.
160-169: LGTM!The multi-stream initialization is well-guarded with a check for
AuxStreamType.Attentionin the dictionary. The fallback toNoneensures graceful degradation to sequential execution when multi-stream is not configured.
171-241: Verify state update thread-safety in multi-stream execution.The
_prefill_forwardmethod updatesssm_statesat line 239 using indexed assignment. When running in parallel with_decode_forward, both methods access the sharedssm_statestensor. This should be safe since:
state_indices_pandstate_indices_dare non-overlapping (split fromstate_indices)- Indexed writes to different memory locations should not conflict
However, please verify that CUDA semantics guarantee atomicity of indexed tensor updates across streams, and that no synchronization issues arise from the shared tensor base.
243-295: LGTM!The
_decode_forwardmethod correctly encapsulates decode computation. Theselective_state_updatecall usesstate_indices_dwhich are disjoint from prefill indices, ensuring safe concurrent execution.
334-397: LGTM!The multi-stream execution logic is well-structured:
- Multi-stream is only enabled when both prefill and decode requests are present, maximizing parallelization benefit
- The sequential fallback handles cases with only one type of request
- Output concatenation maintains correct token ordering (prefill followed by decode)
341-366: Well-structured closure pattern for parallel execution.The closures properly capture the required tensors and delegate to the helper methods. This pattern is consistent with the usage in
NemotronHMOEfor shared/routed expert computation.
Signed-off-by: Wanli Jiang <35160485+Wanli-Jiang@users.noreply.github.com>
Wanli-Jiang
force-pushed
the
user/williamj/support-multistream-mamba-pd
branch
from
3e16819 to
4ca9fb9
Compare
|
/bot run |
|
PR_Github #30271 [ run ] triggered by Bot. Commit: |
|
PR_Github #30271 [ run ] completed with state
|
2 participants
Summary by CodeRabbit
Release Notes
✏️ Tip: You can customize this high-level summary in your review settings.
Description
Test Coverage
PR Checklist
Please review the following before submitting your PR:
PR description clearly explains what and why. If using CodeRabbit's summary, please make sure it makes sense.
PR Follows TRT-LLM CODING GUIDELINES to the best of your knowledge.
Test cases are provided for new code paths (see test instructions)
Any new dependencies have been scanned for license and vulnerabilities
CODEOWNERS updated if ownership changes
Documentation updated as needed
Update tava architecture diagram if there is a significant design change in PR.
The reviewers assigned automatically/manually are appropriate for the PR.
Please check this after reviewing the above items as appropriate for this PR.
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.
Details
run [--reuse-test (optional)pipeline-id --disable-fail-fast --skip-test --stage-list "A10-PyTorch-1, xxx" --gpu-type "A30, H100_PCIe" --test-backend "pytorch, cpp" --add-multi-gpu-test --only-multi-gpu-test --disable-multi-gpu-test --post-merge --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx" --detailed-log --debug(experimental)]Launch build/test pipelines. All previously running jobs will be killed.
--reuse-test (optional)pipeline-id(OPTIONAL) : Allow the new pipeline to reuse build artifacts and skip successful test stages from a specified pipeline or the last pipeline if no pipeline-id is indicated. If the Git commit ID has changed, this option will be always ignored. The DEFAULT behavior of the bot is to reuse build artifacts and successful test results from the last pipeline.--disable-reuse-test(OPTIONAL) : Explicitly prevent the pipeline from reusing build artifacts and skipping successful test stages from a previous pipeline. Ensure that all builds and tests are run regardless of previous successes.--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-PyTorch-1, xxx"(OPTIONAL) : Only run the specified test stages. Examples: "A10-PyTorch-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.--test-backend "pytorch, cpp"(OPTIONAL) : Skip test stages which don't match the specified backends. Only support [pytorch, cpp, tensorrt, triton]. Examples: "pytorch, cpp" (does not run test stages with tensorrt or triton backend). Note: Does NOT update GitHub pipeline 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 in addition to running 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-TensorRT-Post-Merge-1, xxx"(OPTIONAL) : Run the ordinary L0 pre-merge pipeline and specified test stages. Examples: --extra-stage "H100_PCIe-TensorRT-Post-Merge-1, xxx".--detailed-log(OPTIONAL) : Enable flushing out all logs to the Jenkins console. This will significantly increase the log volume and may slow down the job.--debug(OPTIONAL) : Experimental feature. Enable access to the CI container for debugging purpose. Note: Specify exactly one stage in thestage-listparameter to access the appropriate container environment. Note: Does NOT update GitHub check status.For guidance on mapping tests to stage names, see
docs/source/reference/ci-overview.mdand the
scripts/test_to_stage_mapping.pyhelper.kill
killKill all running builds associated with pull request.
skip
skip --comment COMMENTSkip 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-pipelineReuse 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.