[TRTLLM-9676][fix] Fix mamba_cache_manager when enabling cuda_graph_padding and let test cover this case#9873
Conversation
Signed-off-by: jiant <107457950+JadoTu@users.noreply.github.com>
JadoTu
requested review from
Wanli-Jiang,
byshiue,
danielafrimi,
nv-guomingz and
yuantailing
|
/bot run |
📝 WalkthroughWalkthroughThree files are modified to support Mamba state reordering during request padding: cuda_graph_runner.py adds conditional calls to new reordering logic, mamba_cache_manager.py introduces state index reordering handling with request masking, and test configuration is updated with dynamic batch sizes and sampling parameter adjustments. Changes
Estimated code review effort🎯 3 (Moderate) | ⏱️ ~25 minutes
Pre-merge checks and finishing touches❌ Failed checks (1 warning, 1 inconclusive)
✅ 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/pyexecutor/mamba_cache_manager.py (1)
113-120: Reordering logic for padded Mamba state indices looks sound; consider clearing new buffers on shutdownThe new
request_mask/state_indices_arangehelpers andreorder_state_indices_when_padding_requests()correctly ensure that padded slots instate_indicesare filled with distinct, currently unused indices, avoiding repetitions that can upset the CUDA kernels. Under the existing invariants (padded batch size ≤max_batch_size, one block per active request), you will always have at leastpadding_sizefree indices, so the boolean‑mask / advanced indexing pattern is safe.One small lifecycle nit:
shutdown()currently only clearsconv_states,ssm_states, andstate_indices; for completeness and to keep GPU memory hygiene consistent, it would be good to also clearrequest_maskandstate_indices_arangethere.For example:
def shutdown(self): # release tensor memory, keeping python references as tensors self.conv_states = torch.tensor([]) self.ssm_states = torch.tensor([]) self.state_indices = torch.tensor([]) + self.request_mask = torch.tensor([]) + self.state_indices_arange = torch.tensor([]) torch.cuda.empty_cache()Also applies to: 137-147, 179-184
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (1)
393-397: Verify call order: Mamba state reordering must happen afterprepare_resourcesfor the current batchThe idea to special‑case Mamba by calling
reorder_state_indices_when_padding_requests(batch_size, padding_size)when padding is enabled is exactly what’s needed to avoid duplicate SSM state indices for the padded dummy requests.However, this relies on
self.state_indicesalready reflecting the current batch (including the dummy request ID) as populated byMambaCacheManager._prepare_mamba_cache_blocks().From the visible code,
_get_padded_batch()is invoked frompad_batch()before yielding control back to the caller, andMambaCacheManager.prepare_resources()is (in most architectures) called later as part of the main engine flow. If that’s the case,_prepare_mamba_cache_blocks()will overwritestate_indicesafter your reorder call, so duplicates for this step’s padding would remain.You might want to double‑check the actual call order in the engine, and, if
prepare_resources()does indeed run afterpad_batch, consider moving the reordering to a point that is guaranteed to run post‑_prepare_mamba_cache_blocks(for example, a small post‑processing hook insideMambaCacheManager.prepare_resources, guarded by the samepadding_sizelogic passed in via the caller).Please verify the engine’s call sequence to ensure this fix is actually taking effect for the padded CUDA‑graph batches.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (3)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py(2 hunks)tensorrt_llm/_torch/pyexecutor/mamba_cache_manager.py(2 hunks)tests/integration/defs/accuracy/test_llm_api_pytorch.py(2 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/pyexecutor/cuda_graph_runner.pytests/integration/defs/accuracy/test_llm_api_pytorch.pytensorrt_llm/_torch/pyexecutor/mamba_cache_manager.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/pyexecutor/cuda_graph_runner.pytests/integration/defs/accuracy/test_llm_api_pytorch.pytensorrt_llm/_torch/pyexecutor/mamba_cache_manager.py
🧠 Learnings (10)
📚 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/cuda_graph_runner.pytensorrt_llm/_torch/pyexecutor/mamba_cache_manager.py
📚 Learning: 2025-08-08T04:10:19.038Z
Learnt from: djns99
Repo: NVIDIA/TensorRT-LLM PR: 6728
File: cpp/tensorrt_llm/plugins/mixtureOfExperts/mixtureOfExpertsPlugin.cpp:966-966
Timestamp: 2025-08-08T04:10:19.038Z
Learning: TensorRT plugins currently don't support padding functionality, and TensorRT is not getting new features (in maintenance mode). This means that duplicating parameters like mExpertHiddenSize in function calls, even with TODO comments, can be acceptable as pragmatic solutions within these constraints.
Applied to files:
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py
📚 Learning: 2025-07-28T17:06:08.621Z
Learnt from: moraxu
Repo: NVIDIA/TensorRT-LLM PR: 6303
File: tests/integration/test_lists/qa/examples_test_list.txt:494-494
Timestamp: 2025-07-28T17:06:08.621Z
Learning: In TensorRT-LLM testing, it's common to have both CLI flow tests (test_cli_flow.py) and PyTorch API tests (test_llm_api_pytorch.py) for the same model. These serve different purposes: CLI flow tests validate the traditional command-line workflow, while PyTorch API tests validate the newer LLM API backend. Both are legitimate and should coexist.
Applied to files:
tests/integration/defs/accuracy/test_llm_api_pytorch.py
📚 Learning: 2025-08-26T09:49:04.956Z
Learnt from: pengbowang-nv
Repo: NVIDIA/TensorRT-LLM PR: 7192
File: tests/integration/test_lists/test-db/l0_dgx_b200.yml:56-72
Timestamp: 2025-08-26T09:49:04.956Z
Learning: In TensorRT-LLM test configuration files, the test scheduling system handles wildcard matching with special rules that prevent duplicate test execution even when the same tests appear in multiple yaml files with overlapping GPU wildcards (e.g., "*b200*" and "*gb200*").
Applied to files:
tests/integration/defs/accuracy/test_llm_api_pytorch.py
📚 Learning: 2025-09-04T07:33:10.618Z
Learnt from: MrGeva
Repo: NVIDIA/TensorRT-LLM PR: 7219
File: tensorrt_llm/_torch/auto_deploy/compile/backends/torch_cudagraph.py:162-168
Timestamp: 2025-09-04T07:33:10.618Z
Learning: When users explicitly provide cuda_graph_batch_sizes in TorchCudagraphCompiler, respect their choices and only sanitize the values (clamp, dedupe, sort) without forcing additional batch sizes like 1 or max_batch_size. Only add commonly-used batch sizes when falling back to the heuristic.
Applied to files:
tests/integration/defs/accuracy/test_llm_api_pytorch.py
📚 Learning: 2025-09-09T09:40:45.658Z
Learnt from: fredricz-20070104
Repo: NVIDIA/TensorRT-LLM PR: 7645
File: tests/integration/test_lists/qa/llm_function_core.txt:648-648
Timestamp: 2025-09-09T09:40:45.658Z
Learning: In TensorRT-LLM test lists, it's common and intentional for the same test to appear in multiple test list files when they serve different purposes (e.g., llm_function_core.txt for comprehensive core functionality testing and llm_function_core_sanity.txt for quick sanity checks). This duplication allows tests to be run in different testing contexts.
Applied to files:
tests/integration/defs/accuracy/test_llm_api_pytorch.py
📚 Learning: 2025-08-26T09:37:10.463Z
Learnt from: jiaganc
Repo: NVIDIA/TensorRT-LLM PR: 7031
File: tensorrt_llm/bench/dataclasses/configuration.py:90-104
Timestamp: 2025-08-26T09:37:10.463Z
Learning: In TensorRT-LLM's bench configuration, the `get_pytorch_perf_config()` method returns `self.pytorch_config` which is a Dict[str, Any] that can contain default values including `cuda_graph_config`, making the fallback `llm_args["cuda_graph_config"]` safe to use.
Applied to files:
tests/integration/defs/accuracy/test_llm_api_pytorch.py
📚 Learning: 2025-08-26T09:37:10.463Z
Learnt from: jiaganc
Repo: NVIDIA/TensorRT-LLM PR: 7031
File: tensorrt_llm/bench/dataclasses/configuration.py:90-104
Timestamp: 2025-08-26T09:37:10.463Z
Learning: In TensorRT-LLM, the `get_pytorch_perf_config()` method returns `self.pytorch_config` which can contain default `cuda_graph_config` values, so `llm_args` may already have this config before the extra options processing.
Applied to files:
tests/integration/defs/accuracy/test_llm_api_pytorch.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.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
🧬 Code graph analysis (2)
tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (1)
tensorrt_llm/_torch/pyexecutor/mamba_cache_manager.py (3)
MambaCacheManager(28-184)MambaHybridCacheManager(187-268)reorder_state_indices_when_padding_requests(138-147)
tests/integration/defs/accuracy/test_llm_api_pytorch.py (3)
tensorrt_llm/llmapi/llm_args.py (1)
CudaGraphConfig(102-159)tests/integration/defs/accuracy/accuracy_core.py (1)
GSM8K(334-349)tensorrt_llm/evaluate/lm_eval.py (1)
GSM8K(455-510)
⏰ 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)
tests/integration/defs/accuracy/test_llm_api_pytorch.py (2)
4539-4544: Explicit CUDA graphbatch_sizeslist looks appropriateUsing an explicit, sorted
batch_sizeslist withenable_padding=Truewhencuda_graphis enabled is consistent with howCudaGraphConfigis intended to be used and should exercise the padding path over a wide range of batch sizes. No issues from my side here.
4559-4559: GSM8KNUM_SAMPLESpatch is safe and test‑localizes behaviorPatching
GSM8K.NUM_SAMPLESto 1319 inside the test keeps this test independent from any global modifications of that attribute elsewhere and aligns with the full‑dataset default. Looks good.tensorrt_llm/_torch/pyexecutor/cuda_graph_runner.py (1)
18-18: Mamba cache manager import is appropriate and localizedImporting
MambaCacheManager/MambaHybridCacheManagerhere for type checks in_get_padded_batchis reasonable and doesn’t introduce obvious cyclic dependencies.
|
PR_Github #27702 [ run ] triggered by Bot. Commit: |
|
/bot run |
|
PR_Github #27949 [ run ] triggered by Bot. Commit: |
|
PR_Github #27949 [ run ] completed with state |
Signed-off-by: JadoTu <107457950+JadoTu@users.noreply.github.com>
|
/bot run |
|
PR_Github #28326 [ run ] triggered by Bot. Commit: |
|
PR_Github #28326 [ run ] completed with state |
Signed-off-by: jiant <107457950+JadoTu@users.noreply.github.com>
|
/bot run |
|
PR_Github #29709 [ run ] triggered by Bot. Commit: |
|
PR_Github #29709 [ run ] completed with state
|
|
/bot run |
Signed-off-by: jiant <107457950+JadoTu@users.noreply.github.com>
|
/bot run |
|
PR_Github #30428 [ run ] triggered by Bot. Commit: |
|
PR_Github #30428 [ run ] completed with state
|
|
/bot run |
|
PR_Github #30435 [ run ] triggered by Bot. Commit: |
|
PR_Github #30435 [ run ] completed with state
|
|
/bot run |
|
PR_Github #30464 [ run ] triggered by Bot. Commit: |
|
PR_Github #30464 [ run ] completed with state |
Signed-off-by: jiant <107457950+JadoTu@users.noreply.github.com>
|
Commit 2ecd8a3 is a intermediate product as it solves the cuda D2H sync but brings another CuBin kernel launch overhead. |
Signed-off-by: jiant <107457950+JadoTu@users.noreply.github.com>
|
Commit ce52308 finally solve the runtime launch overhead. |
|
/bot run |
|
PR_Github #30711 [ run ] triggered by Bot. Commit: |
|
PR_Github #30711 [ run ] completed with state
|
|
/bot run |
|
PR_Github #30757 [ run ] triggered by Bot. Commit: |
|
PR_Github #30757 [ run ] completed with state |
syuoni
left a comment
syuoni
left a comment
There was a problem hiding this comment.
Approved for the cuda graph and test changes.
…adding and let test cover this case (NVIDIA#9873) Signed-off-by: JadoTu <107457950+JadoTu@users.noreply.github.com> Signed-off-by: Daniil Kulko <kulkodaniil@gmail.com>
5 participants
Summary by CodeRabbit
Improvements
Tests
✏️ 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.