[TRTLLM-8827] [feat] Enable low precision alltoall for Cutlass and TRTLLMGen backends #8675
Conversation
kaiyux
requested review from
Naveassaf,
QiJune,
bobboli,
dongxuy04,
xxi-nv and
zongfeijing
📝 WalkthroughWalkthroughThe changes centralize the Changes
Sequence DiagramsequenceDiagram
participant Init as Module __init__
participant Env as Environment
participant Select as select_alltoall_method_type()
participant Model as model_config
participant Memory as MnnvlMemory
participant Forward as forward_*()
participant Kernel as fused_moe_kernel
Init->>Env: Check TRTLLM_FORCE_ALLTOALL_METHOD
Init->>Select: Call method selector
Select->>Env: Fetch override if set
Select->>Model: Read routing/combine conditions
Select->>Select: Determine type (MNNVL/DeepEP/etc.)
Select-->>Init: Return alltoall_method_type
alt Method Type = MNNVL
Init->>Memory: Initialize MnnvlMemory
Memory-->>Init: Provide workspaces
Init->>Model: Read use_low_precision_moe_combine
Init-->>Init: Set use_low_precision_combine flag
else Method Type = DeepEP/DeepEPLowLatency
Init-->>Init: Raise NotImplementedError
else Other/Unknown
Init-->>Init: Raise NotImplementedError
end
Forward->>Kernel: Pass use_low_precision_combine flag
Kernel-->>Forward: Execute with precision setting
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~20 minutes
Pre-merge checks and finishing touches❌ Failed checks (2 warnings)
✅ Passed checks (1 passed)
✨ Finishing touches
🧪 Generate unit tests (beta)
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: 4
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (3)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_wide_ep.py (1)
1-5: Add NVIDIA Apache-2.0 header (2025) at file topRequired by coding guidelines; currently missing.
(Use the same header block shown in utils.py.)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py (1)
1-10: Add NVIDIA Apache-2.0 header (2025) at file topRequired by coding guidelines; currently missing.
(Use the same header block shown in utils.py.)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py (1)
1-10: Add NVIDIA Apache-2.0 header (2025) at file topRequired by coding guidelines; currently missing.
(Use the same header block shown in utils.py.)
🧹 Nitpick comments (7)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_wide_ep.py (3)
175-181: Log key should be class-specific to avoid suppressionUsing the same info_once key across classes can hide messages. Prefer a per-class key.
- logger.info_once( - f"{self.__class__.__name__} selects alltoall_method_type {self.alltoall_method_type!r}", - key="alltoall_method_type") + logger.info_once( + f"{self.__class__.__name__} selects alltoall_method_type {self.alltoall_method_type!r}", + key=f"{self.__class__.__name__}.alltoall_method_type")
247-296: Harden env override parsing; avoid KeyError on typosAlltoallMethodType[...] raises KeyError for invalid TRTLLM_FORCE_ALLTOALL_METHOD values. Catch and log for safer behavior.
- all2all_method_type = os.environ.get("TRTLLM_FORCE_ALLTOALL_METHOD") - if all2all_method_type is not None: - return AlltoallMethodType[all2all_method_type] + all2all_method_type = os.environ.get("TRTLLM_FORCE_ALLTOALL_METHOD") + if all2all_method_type is not None: + try: + return AlltoallMethodType[all2all_method_type] + except KeyError: + logger.warning( + f"Invalid TRTLLM_FORCE_ALLTOALL_METHOD='{all2all_method_type}'. " + f"Valid: {[e.name for e in AlltoallMethodType]}. Falling back to auto selection.")
330-358: Potentially heavy mask allocation in low-latency adapterThe expert-by-token boolean mask can be large. Consider computing indices without materializing a full mask to reduce memory traffic.
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py (2)
145-171: Env override parsing: guard against invalid valuesSame concern as WideEPMoE; catch KeyError and warn to avoid hard crashes on typos.
- all2all_method_type = os.environ.get("TRTLLM_FORCE_ALLTOALL_METHOD") - if all2all_method_type is not None: - if AlltoallMethodType[all2all_method_type] in [ + all2all_method_type = os.environ.get("TRTLLM_FORCE_ALLTOALL_METHOD") + if all2all_method_type is not None: + try: + forced = AlltoallMethodType[all2all_method_type] + except KeyError: + logger.warning( + f"Invalid TRTLLM_FORCE_ALLTOALL_METHOD='{all2all_method_type}'. " + f"Valid: {[e.name for e in AlltoallMethodType]}. Falling back to auto selection.") + else: + if forced in [ AlltoallMethodType.DeepEP, AlltoallMethodType.DeepEPLowLatency - ]: - raise NotImplementedError( - "DeepEP and DeepEPLowLatency are not supported for CutlassFusedMoE yet" - ) - return AlltoallMethodType[all2all_method_type] + ]: + raise NotImplementedError( + "DeepEP and DeepEPLowLatency are not supported for TRTLLMGenFusedMoE yet" + ) + return forced
117-119: Unique logger key to avoid cross-class suppressionUse a class-specific key.
- logger.info_once( - f"{self.__class__.__name__} selects alltoall_method_type {self.alltoall_method_type!r}", - key="alltoall_method_type") + logger.info_once( + f"{self.__class__.__name__} selects alltoall_method_type {self.alltoall_method_type!r}", + key=f"{self.__class__.__name__}.alltoall_method_type")tensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py (2)
147-149: Unique logger key to avoid cross-class suppressionUse a class-specific key.
- logger.info_once( - f"{self.__class__.__name__} selects alltoall_method_type {self.alltoall_method_type!r}", - key="alltoall_method_type") + logger.info_once( + f"{self.__class__.__name__} selects alltoall_method_type {self.alltoall_method_type!r}", + key=f"{self.__class__.__name__}.alltoall_method_type")
211-237: Harden env override parsing; avoid KeyError on typosSame suggestion as in TRTLLMGen/WideEPMoE.
- all2all_method_type = os.environ.get("TRTLLM_FORCE_ALLTOALL_METHOD") - if all2all_method_type is not None: - if AlltoallMethodType[all2all_method_type] in [ + all2all_method_type = os.environ.get("TRTLLM_FORCE_ALLTOALL_METHOD") + if all2all_method_type is not None: + try: + forced = AlltoallMethodType[all2all_method_type] + except KeyError: + logger.warning( + f"Invalid TRTLLM_FORCE_ALLTOALL_METHOD='{all2all_method_type}'. " + f"Valid: {[e.name for e in AlltoallMethodType]}. Falling back to auto selection.") + else: + if forced in [ AlltoallMethodType.DeepEP, AlltoallMethodType.DeepEPLowLatency - ]: - raise NotImplementedError( - "DeepEP and DeepEPLowLatency are not supported for CutlassFusedMoE yet" - ) - return AlltoallMethodType[all2all_method_type] + ]: + raise NotImplementedError( + "DeepEP and DeepEPLowLatency are not supported for CutlassFusedMoE yet" + ) + return forced
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (4)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py(4 hunks)tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py(3 hunks)tensorrt_llm/_torch/modules/fused_moe/fused_moe_wide_ep.py(1 hunks)tensorrt_llm/_torch/utils.py(2 hunks)
🧰 Additional context used
📓 Path-based instructions (3)
**/*.{h,hpp,hh,hxx,cpp,cxx,cc,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Use only spaces, no tabs; indent with 4 spaces.
Files:
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.pytensorrt_llm/_torch/modules/fused_moe/fused_moe_wide_ep.pytensorrt_llm/_torch/utils.pytensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: Python code must target Python 3.8+.
Indent Python code with 4 spaces; do not use tabs.
Maintain module namespace when importing; prefer 'from package.subpackage import foo' then 'foo.SomeClass()' instead of importing the class directly.
Python filenames should be snake_case (e.g., some_file.py).
Python classes use PascalCase names.
Functions and methods use snake_case names.
Local variables use snake_case; prefix 'k' for variables that start with a number (e.g., k_99th_percentile).
Global variables use upper SNAKE_CASE prefixed with 'G' (e.g., G_MY_GLOBAL).
Constants use upper SNAKE_CASE (e.g., MY_CONSTANT).
Avoid shadowing variables from an outer scope.
Initialize all externally visible members of a class in the constructor.
Prefer docstrings for interfaces that may be used outside a file; comments for in-function or file-local interfaces.
Use Google-style docstrings for classes and functions (Sphinx-parsable).
Document attributes and variables inline so they render under the class/function docstring.
Avoid reflection when a simpler, explicit approach suffices (e.g., avoid dict(**locals()) patterns).
In try/except, catch the most specific exceptions possible.
For duck-typing try/except, keep the try body minimal and use else for the main logic.
Files:
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.pytensorrt_llm/_torch/modules/fused_moe/fused_moe_wide_ep.pytensorrt_llm/_torch/utils.pytensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py
**/*.{cpp,cxx,cc,h,hpp,hh,hxx,cu,cuh,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
Prepend the NVIDIA Apache-2.0 copyright header with current year to the top of all source files (e.g., .cpp, .h, .cu, .py).
Files:
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.pytensorrt_llm/_torch/modules/fused_moe/fused_moe_wide_ep.pytensorrt_llm/_torch/utils.pytensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py
🧬 Code graph analysis (3)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py (6)
tensorrt_llm/_torch/custom_ops/trtllm_gen_custom_ops.py (1)
fp4_block_scale_fake_output_without_finalize(308-331)tensorrt_llm/_torch/utils.py (2)
AlltoallMethodType(320-328)Fp4QuantizedTensor(99-106)tensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py (2)
select_alltoall_method_type(211-236)enable_alltoall(239-242)tensorrt_llm/_torch/modules/fused_moe/fused_moe_wide_ep.py (2)
select_alltoall_method_type(248-296)enable_alltoall(305-308)tensorrt_llm/logger.py (1)
info_once(141-142)tensorrt_llm/_mnnvl_utils.py (4)
MnnvlMemory(53-338)MnnvlMoe(352-624)get_moe_workspaces(360-376)supports_mnnvl(332-338)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_wide_ep.py (1)
tensorrt_llm/_torch/utils.py (2)
AlltoallMethodType(320-328)Fp4QuantizedTensor(99-106)
tensorrt_llm/_torch/modules/fused_moe/fused_moe_cutlass.py (4)
tensorrt_llm/_torch/utils.py (1)
AlltoallMethodType(320-328)tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py (2)
select_alltoall_method_type(145-170)enable_alltoall(173-176)tensorrt_llm/logger.py (1)
info_once(141-142)tensorrt_llm/_mnnvl_utils.py (6)
MnnvlMemory(53-338)initialize(91-100)MnnvlMoe(352-624)get_moe_workspaces(360-376)get_moe_prepare_workspace(379-390)supports_mnnvl(332-338)
⏰ 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 (3)
tensorrt_llm/_torch/utils.py (1)
319-328: Centralized AlltoallMethodType looks goodEnum values are explicit and stable for cross-module use. No issues found.
tensorrt_llm/_torch/modules/fused_moe/fused_moe_wide_ep.py (1)
199-211: Confirm NVSHMEM_QP_DEPTH timing relative to NVSHMEM initializationSetting NVSHMEM_QP_DEPTH after NVSHMEM initializes may be ignored. Ensure this executes before any NVSHMEM init in the process.
Please verify initialization order in runtime setup; if needed, document and/or set this earlier (e.g., in a launcher).
tensorrt_llm/_torch/modules/fused_moe/fused_moe_trtllm_gen.py (1)
114-119: Blocker: select_alltoall_method_type call signature mismatchMethod takes only self, but call passes 4 args -> runtime TypeError.
Apply this diff:
- self.alltoall_method_type = self.select_alltoall_method_type( - model_config.mapping, routing_method.experts_per_token, dtype, - model_config.use_cuda_graph) + self.alltoall_method_type = self.select_alltoall_method_type()Likely an incorrect or invalid review comment.
|
/bot run |
|
PR_Github #22632 [ run ] triggered by Bot. Commit: |
|
PR_Github #22632 [ run ] completed with state |
|
/bot run |
|
PR_Github #22665 [ run ] triggered by Bot. Commit: |
|
PR_Github #22665 [ run ] completed with state |
Signed-off-by: Kaiyu Xie <[email protected]>
Signed-off-by: Kaiyu Xie <[email protected]>
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com> Signed-off-by: Kaiyu Xie <[email protected]>
Signed-off-by: Kaiyu Xie <[email protected]>
kaiyux
force-pushed
the
user/kaiyu/alltoall
branch
from
45784e3 to
1ffaaae
Compare
Signed-off-by: Kaiyu Xie <[email protected]>
|
/bot run --disable-fail-fast |
|
PR_Github #22764 [ run ] triggered by Bot. Commit: |
|
PR_Github #22764 [ run ] completed with state |
…TLLMGen backends (NVIDIA#8675) Signed-off-by: Kaiyu Xie <[email protected]> Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
…TLLMGen backends (NVIDIA#8675) Signed-off-by: Kaiyu Xie <[email protected]> Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
…TLLMGen backends (NVIDIA#8675) Signed-off-by: Kaiyu Xie <[email protected]> Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
…TLLMGen backends (NVIDIA#8675) Signed-off-by: Kaiyu Xie <[email protected]> Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
6 participants
Summary by CodeRabbit
Release Notes
New Features
Refactor
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
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.
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.