[TRTLLM-9432][feat] Reduce synchronization and recompilation for qwen3-next - #9691
Conversation
📝 WalkthroughWalkthroughThis pull request refactors query sequence length handling in Mamba2 attention mechanisms by replacing Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 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: 0
🧹 Nitpick comments (1)
tensorrt_llm/_torch/pyexecutor/mamba_cache_manager.py (1)
126-128: Consider optimizing the tensor copy operation.Creating a pinned CPU tensor and then copying to GPU adds overhead. Since
self.state_indicesis already on the CUDA device, consider directly creating the tensor on the target device:Apply this diff for a more efficient approach:
- self.state_indices[:len(state_indices)].copy_(torch.tensor( - state_indices, dtype=torch.int32, pin_memory=True), - non_blocking=True) + self.state_indices[:len(state_indices)].copy_( + torch.tensor(state_indices, dtype=torch.int32, device=self.state_indices.device), + non_blocking=False)Alternatively, if the intent is to support async transfers from CPU, ensure
state_indicesis on CPU and the copy pattern makes sense in the broader context.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (4)
tensorrt_llm/_torch/models/modeling_qwen3_next.py(6 hunks)tensorrt_llm/_torch/modules/fla/l2norm.py(1 hunks)tensorrt_llm/_torch/modules/mamba/mamba2_metadata.py(3 hunks)tensorrt_llm/_torch/pyexecutor/mamba_cache_manager.py(1 hunks)
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
**/*.py: The 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 (e.g., usefrom package.subpackage import fooand thenfoo.SomeClass()instead offrom package.subpackage.foo import SomeClass)
Python filenames should use snake_case (e.g.,some_file.py)
Python class names should use PascalCase (e.g.,class SomeClass)
Python function and method names should use snake_case (e.g.,def my_awesome_function():)
Python local variable names should use snake_case, with prefixkfor variable names that start with a number (e.g.,k_99th_percentile = ...)
Python global variables should use upper snake_case with prefixG(e.g.,G_MY_GLOBAL = ...)
Python constants should use upper snake_case (e.g.,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 for Python classes and functions, which can be parsed by Sphinx
Python attributes and variables can be documented inline with type and description (e.g.,self.x = 5followed by"""<type>: Description of 'x'""")
Avoid using reflection in Python when functionality can be easily achieved without reflection
When using try-except blocks in Python, limit the except clause to the smallest set of specific errors possible instead of catching all exceptions
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 and use the else block to implement the logic
Files:
tensorrt_llm/_torch/modules/fla/l2norm.pytensorrt_llm/_torch/modules/mamba/mamba2_metadata.pytensorrt_llm/_torch/pyexecutor/mamba_cache_manager.pytensorrt_llm/_torch/models/modeling_qwen3_next.py
**/*.{cpp,h,cu,py}
📄 CodeRabbit inference engine (CODING_GUIDELINES.md)
All TensorRT-LLM Open Source Software code files should contain an NVIDIA copyright header that includes the current year at the top
Files:
tensorrt_llm/_torch/modules/fla/l2norm.pytensorrt_llm/_torch/modules/mamba/mamba2_metadata.pytensorrt_llm/_torch/pyexecutor/mamba_cache_manager.pytensorrt_llm/_torch/models/modeling_qwen3_next.py
🧠 Learnings (8)
📚 Learning: 2025-10-20T16:54:09.824Z
Learnt from: nvchenghaoz
Repo: NVIDIA/TensorRT-LLM PR: 8469
File: tensorrt_llm/_torch/auto_deploy/custom_ops/rms_norm.py:6-6
Timestamp: 2025-10-20T16:54:09.824Z
Learning: In tensorrt_llm/_torch/auto_deploy/custom_ops/rms_norm.py, the import `from ...modules.mamba.layernorm_gated import _layer_norm_fwd` is correct and should not be changed to modules.fla.layernorm_gated. The _layer_norm_fwd function exists in both modules/mamba/layernorm_gated.py and modules/fla/layernorm_gated.py, but the mamba version is the intended implementation for this use case.
Applied to files:
tensorrt_llm/_torch/modules/fla/l2norm.py
📚 Learning: 2025-10-20T17:09:21.560Z
Learnt from: nvchenghaoz
Repo: NVIDIA/TensorRT-LLM PR: 8469
File: tensorrt_llm/_torch/auto_deploy/transform/library/rms_norm.py:180-182
Timestamp: 2025-10-20T17:09:21.560Z
Learning: In tensorrt_llm/_torch/auto_deploy/transform/library/rms_norm.py, the _gated_rmsnorm_replacement function does not need to cast the output of torch.ops.auto_deploy.torch_rmsnorm_gated back to the input dtype, even though the custom op returns fp32. The dtype handling is managed elsewhere or the fp32 output is acceptable for downstream consumers.
Applied to files:
tensorrt_llm/_torch/modules/fla/l2norm.py
📚 Learning: 2025-08-19T12:45:11.997Z
Learnt from: amitz-nv
Repo: NVIDIA/TensorRT-LLM PR: 7033
File: tensorrt_llm/_torch/pyexecutor/model_engine.py:0-0
Timestamp: 2025-08-19T12:45:11.997Z
Learning: In tensorrt_llm/_torch/pyexecutor/model_engine.py, DoRA (Delta Orthogonal Rank Adaptation) functionality was removed from the PyTorch flow to eliminate issues with inverted DoRA detection logic. The original is_dora condition was checking if scaling_vec_pointer == 0, which was potentially incorrect.
Applied to files:
tensorrt_llm/_torch/modules/fla/l2norm.py
📚 Learning: 2025-09-23T15:12:38.312Z
Learnt from: nv-lschneider
Repo: NVIDIA/TensorRT-LLM PR: 7910
File: cpp/tensorrt_llm/thop/allreduceOp.cpp:352-446
Timestamp: 2025-09-23T15:12:38.312Z
Learning: In TensorRT-LLM NCCL device implementation, NCCL version 2.28+ requirements are handled at runtime in the nccl_device/config layer rather than with compile-time guards. This allows the allreduceOp to remain version-agnostic and delegates version compatibility validation to the appropriate lower-level components that can gracefully handle unsupported configurations.
Applied to files:
tensorrt_llm/_torch/modules/fla/l2norm.py
📚 Learning: 2025-08-21T02:39:12.009Z
Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 7104
File: cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu:1475-1480
Timestamp: 2025-08-21T02:39:12.009Z
Learning: The min latency mode functionality in TensorRT-LLM MOE kernels (cpp/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_kernels.cu) is deprecated and no longer being maintained/updated, as confirmed by djns99. Bug reports and optimization suggestions for the computeStridesTmaWarpSpecializedLowLatencyKernel and related min latency code paths should be deprioritized.
Applied to files:
tensorrt_llm/_torch/modules/fla/l2norm.py
📚 Learning: 2025-08-15T06:46:54.897Z
Learnt from: eopXD
Repo: NVIDIA/TensorRT-LLM PR: 6767
File: cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp:0-0
Timestamp: 2025-08-15T06:46:54.897Z
Learning: In cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp addToken function, newly allocated blocks are unshared by design. The beam search path in addToken (when sequence.getNumTokens() > windowSize) is currently broken/non-functional with SWA, so the block allocation doesn't follow a shared-then-unshared pattern.
Applied to files:
tensorrt_llm/_torch/pyexecutor/mamba_cache_manager.py
📚 Learning: 2025-08-21T09:41:49.347Z
Learnt from: eopXD
Repo: NVIDIA/TensorRT-LLM PR: 6768
File: cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp:2010-2045
Timestamp: 2025-08-21T09:41:49.347Z
Learning: In cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp, updateSequenceCacheBlockOffsets is specifically for updating bookkeeping when blocks are added during the context phase, not for refreshing offsets after detach operations. During detach operations, GenerationRequest::removeFrontBlock handles the necessary cache block bookkeeping internally.
Applied to files:
tensorrt_llm/_torch/pyexecutor/mamba_cache_manager.py
📚 Learning: 2025-08-14T21:04:50.248Z
Learnt from: thorjohnsen
Repo: NVIDIA/TensorRT-LLM PR: 6910
File: cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp:0-0
Timestamp: 2025-08-14T21:04:50.248Z
Learning: In KV cache onboarding logic during prefill in cpp/tensorrt_llm/batch_manager/kvCacheManager.cpp, when calculating which blocks fall within the attention window, use getTokensPerBlock() to advance token indices rather than block->getUniqueTokens().size(), because the calculation needs to consider the post-prefill state where blocks will be filled to capacity, not their current token count.
Applied to files:
tensorrt_llm/_torch/pyexecutor/mamba_cache_manager.pytensorrt_llm/_torch/models/modeling_qwen3_next.py
⏰ 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 (9)
tensorrt_llm/_torch/modules/fla/l2norm.py (1)
55-63: LGTM! Reduced kernel recompilation overhead.Changing
Tfromtl.constexprto a runtime parameter and removing the unusedNBparameter reduces kernel recompilation, which aligns with the PR objective to reduce recompilation overhead.tensorrt_llm/_torch/modules/mamba/mamba2_metadata.py (2)
126-129: LGTM! Clear explanation for dual attribute design.The comment clearly explains why both
query_start_loc(int32) andquery_start_loc_long(long) are needed for different kernel requirements. The implementation correctly handles the type conversion.
149-155: No action required. The code correctly handles the case wherequery_start_locisNone. Downstream code paths that requirequery_start_loccontain explicit guards and assertions that prevent accessing it when it isNone. Settingquery_start_loc_longas an alternative whenquery_start_locisNoneis an intentional design pattern in the Mamba2 implementation, not an inconsistency.Likely an incorrect or invalid review comment.
tensorrt_llm/_torch/models/modeling_qwen3_next.py (6)
824-831: LGTM! Consistent parameter rename.The method signature update from
cu_seqlenstoquery_start_loc_longaligns with the broader refactoring and improves parameter naming clarity.
1015-1017: LGTM! Improved in-place initialization.Using
.zero_()for in-place zeroing is more efficient and clearer than assignment to0. The commented line forconv_statesindicates awareness that it may not be necessary.
889-890: Verify that query_start_loc is not None when used.Line 889 retrieves
query_start_locfrom kwargs, which can beNonewhennum_contexts == 0(per mamba2_metadata.py line 150). Verify that this code path is only executed whennum_prefills > 0(checked at line 1057), ensuringquery_start_locis notNone.The control flow suggests this is safe since
forward_extendis only called whennum_prefills > 0(line 1057-1058), which guaranteesquery_start_locis notNone. However, please confirm this assumption holds in all execution paths.
869-869: Verify parameter compatibility with kernel signature.Ensure that
fused_sigmoid_gating_delta_rule_updateaccepts a parameter namedcu_seqlensthat can be passedquery_start_loc_long(a long tensor). The function definition and parameter type annotations should be checked to confirm this parameter exists and accepts the provided tensor type.
971-971: Verify parameter compatibility with kernel signature.The
cu_seqlensparameter passed tochunk_gated_delta_rulemust accept a long tensor (e.g.,query_start_loc_long). Locate thechunk_gated_delta_rulefunction definition and confirm the parameter name and expected tensor type match.
904-904: Potential index mismatch in slicing at line 904.Line 904 slices
query_start_loc[:num_prefill + 1]. Sincequery_start_locis defined from kwargs and according to mamba2_metadata.py, representscu_seqlens[:batch_size + 1], verify thatnum_prefilldoes not exceed the valid index range. If prefills are a subset of the batch, confirm thatnum_prefill < batch_sizeto ensure the slice[:num_prefill + 1]is within bounds ofquery_start_loc.
|
/bot run |
|
PR_Github #26884 [ run ] triggered by Bot. Commit: |
|
PR_Github #26884 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #26978 [ run ] triggered by Bot. Commit: |
|
PR_Github #26978 [ run ] completed with state |
66dea95 to
527ded0
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #27076 [ run ] triggered by Bot. Commit: |
|
PR_Github #27076 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #27109 [ run ] triggered by Bot. Commit: |
|
PR_Github #27109 [ run ] completed with state |
|
/bot run --disable-fail-fast --add-multi-gpu-test |
|
PR_Github #27149 [ run ] triggered by Bot. Commit: |
|
PR_Github #27149 [ run ] completed with state |
|
/bot run --disable-fail-fast --add-multi-gpu-test |
|
PR_Github #27209 [ run ] triggered by Bot. Commit: |
|
PR_Github #27209 [ run ] completed with state |
8dc049e to
78ac1cc
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #28310 [ run ] triggered by Bot. Commit: |
|
/rebase |
Signed-off-by: Tailing Yuan <yuantailing@gmail.com>
Signed-off-by: Tailing Yuan <yuantailing@gmail.com>
Signed-off-by: Tailing Yuan <yuantailing@gmail.com>
Signed-off-by: Tailing Yuan <yuantailing@gmail.com>
Signed-off-by: Tailing Yuan <yuantailing@gmail.com>
Removed 'num_decodes' parameter from forward_decode method and its usage. Signed-off-by: Tailing Yuan <yuantailing@gmail.com>
8b74a39 to
1f27c84
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #29349 [ run ] triggered by Bot. Commit: |
|
PR_Github #29349 [ run ] completed with state
|
|
/bot run --disable-fail-fast |
|
PR_Github #29392 [ run ] triggered by Bot. Commit: |
|
PR_Github #29392 [ run ] completed with state |
…3-next (NVIDIA#9691) Signed-off-by: Tailing Yuan <yuantailing@gmail.com>
…3-next (NVIDIA#9691) Signed-off-by: Tailing Yuan <yuantailing@gmail.com> Signed-off-by: Daniil Kulko <kulkodaniil@gmail.com>
7 participants
Summary by CodeRabbit
✏️ 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.