[https://nvbugs/5550409][fix] Disable torch compile in piecewise attention part to Avoid host overhead #8708
Conversation
yizhang-nv
requested review from
hyukn,
kaiyux,
liji-nv,
yuantailing and
yuxianq
and removed request for
hyukn and
yuxianq
📝 WalkthroughWalkthroughThis pull request introduces a piecewise runner orchestration system with conditional torch.compile execution. It adds runner lifecycle tracking to PiecewiseInterpreter and PiecewiseRunner, implements first/last runner signaling via set_piecewise_running, and conditionally bypasses torch.compile during piecewise execution in attention modules. Changes
Sequence Diagram(s)sequenceDiagram
participant Optimizer as piecewise_optimizer
participant Interpreter as PiecewiseInterpreter
participant Runner as PiecewiseRunner
participant Utils as utils
Optimizer->>Interpreter: __init__(piecewise_runner_num)
Interpreter->>Interpreter: Initialize piecewise_runner_idx = 0
loop For each module
Interpreter->>Runner: __init__(..., is_first_runner, is_last_runner)
Runner->>Runner: Store is_first_runner, is_last_runner
alt is_first_runner or is_last_runner
Runner->>Utils: set_piecewise_running(True)
end
Interpreter->>Interpreter: piecewise_runner_idx += 1
end
Note over Interpreter,Runner: Runner lifecycle complete
sequenceDiagram
participant Module as attention.forward
participant Compile as maybe_compile wrapper
participant Utils as is_piecewise_running()
participant Torch as torch.compile
Module->>Compile: maybe_compiled_copy_(dst, src)
Compile->>Utils: is_piecewise_running()?
alt Piecewise running (True)
Compile-->>Module: Return original function
Module->>Module: Execute uncompiled
else Normal execution (False)
Compile->>Torch: Apply torch.compile
Torch-->>Compile: Return compiled variant
Compile-->>Module: Return compiled function
Module->>Module: Execute compiled
end
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: 3
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (4)
tensorrt_llm/_torch/utils.py (1)
1-1: Add NVIDIA Apache-2.0 header (2025) to comply with repo policy.
Please prepend the standard NVIDIA Apache-2.0 copyright header with year 2025.As per coding guidelines.
tensorrt_llm/_torch/compilation/piecewise_optimizer.py (2)
1-1: Add NVIDIA Apache-2.0 header (2025).
Required by project policy.As per coding guidelines.
2-2: Python 3.8+ typing: replace 3.10+ syntax and fix tuple/optional types.
The project targets Python 3.8+. Current hints (int | torch.SymInt, list[int], tuple[…], tuple[GraphModule, int]) won’t parse; also runtime_num_tokens_idx is a pair or None.Apply:
-from typing import Callable, List, Optional, Sequence, Union +from typing import Callable, List, Optional, Sequence, Tuple, Union class PiecewiseInterpreter(Interpreter): def __init__( self, module: GraphModule, enable_inductor: bool, - compile_time_num_tokens: Union[int | torch.SymInt], - capture_num_tokens: list[int], - exclude_modules_id: list[int], + compile_time_num_tokens: Union[int, torch.SymInt], + capture_num_tokens: List[int], + exclude_modules_id: List[int], piecewise_runner_num: int, - graph_pool_handle: tuple[int, int], + graph_pool_handle: Tuple[int, int], ... ): class PiecewiseRunner(object): def __init__( self, graph: GraphModule, name: str, - compile_time_num_tokens: Union[int | torch.SymInt], - runtime_num_tokens_idx: tuple[int], + compile_time_num_tokens: Union[int, torch.SymInt], + runtime_num_tokens_idx: Optional[Tuple[int, int]], capture_num_tokens: List[int], graph_pool_handle, default_callable: Callable, enable_inductor: bool, is_first_runner: bool, is_last_runner: bool, ): def piecewise_optimizer( gm: GraphModule, example_inputs: List[torch.Tensor], enable_inductor: bool, - input_num_tokens: Union[int | torch.SymInt], + input_num_tokens: Union[int, torch.SymInt], capture_num_tokens: Sequence[int], - graph_pool_handle: tuple[int, int], + graph_pool_handle: Tuple[int, int], max_num_streams: int = 1, -) -> tuple[GraphModule, int]: +) -> Tuple[GraphModule, int]:Optional cleanups: prefer “is not None” over “!= None”.
Also applies to: 24-36, 123-135, 235-244, 243-244
tensorrt_llm/_torch/modules/attention.py (1)
1-1: Add NVIDIA Apache-2.0 header (2025).
Please add the standard header at file top.As per coding guidelines.
🧹 Nitpick comments (2)
tensorrt_llm/_torch/utils.py (1)
15-15: Naming: consider G_IS_PIECEWISE_RUNNING for globals per style.
If you keep any process‑global flags, prefer G_UPPER_SNAKE_CASE.As per coding guidelines.
tensorrt_llm/_torch/compilation/piecewise_optimizer.py (1)
136-167: None checks and small nits.
Use “is not None” for None checks; rename runtime_num_of_token -> num_tokens_at_runtime for clarity.
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (3)
tensorrt_llm/_torch/compilation/piecewise_optimizer.py(8 hunks)tensorrt_llm/_torch/modules/attention.py(6 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/utils.pytensorrt_llm/_torch/compilation/piecewise_optimizer.pytensorrt_llm/_torch/modules/attention.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/utils.pytensorrt_llm/_torch/compilation/piecewise_optimizer.pytensorrt_llm/_torch/modules/attention.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/utils.pytensorrt_llm/_torch/compilation/piecewise_optimizer.pytensorrt_llm/_torch/modules/attention.py
🧬 Code graph analysis (2)
tensorrt_llm/_torch/compilation/piecewise_optimizer.py (1)
tensorrt_llm/_torch/utils.py (3)
get_piecewise_cuda_graph_flag(278-280)make_weak_ref(91-106)set_piecewise_running(44-46)
tensorrt_llm/_torch/modules/attention.py (1)
tensorrt_llm/_torch/utils.py (2)
is_piecewise_running(49-51)is_torch_compiling(39-41)
⏰ 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 (1)
tensorrt_llm/_torch/modules/attention.py (1)
1231-1235: Verification confirms all oldcompiled_*references have been successfully migrated tomaybe_compiled_*wrappers.The ripgrep search found only the new wrapper function definitions (line 87:
maybe_compiled_copy_, line 92:maybe_compiled_cat) and their usages at lines 1232, 1328, 1423, and 1481—all correctly using themaybe_compiled_*prefix. No legacycompiled_copy_orcompiled_catcalls remain.
Signed-off-by: yizhang-nv <[email protected]>
Signed-off-by: yizhang-nv <[email protected]>
yizhang-nv
force-pushed
the
disable-compile-in-piecewise
branch
from
17bcf22 to
05fff03
Compare
|
/bot run |
|
PR_Github #22736 [ run ] triggered by Bot. Commit: |
|
PR_Github #22736 [ run ] completed with state |
yizhang-nv
changed the title
There was a problem hiding this comment.
Approved according to the following discussion:
Q: Are there any pitfalls with piecewise_runner_idx? For example, could it happen that a larger idx gets called first? Or could the number of call_module calls not equal len(set(node_to_graph_id.values())) - len(exclude_modules_id)?
A: This won't happen, the idx in piecewise is guaranteed to maintain order. len(set(node_to_graph_id.values())) calculates how many graphs were split in total, and len(exclude_modules_id) represents how many subgraphs run under eager mode.
5 participants
This PR disables torch.compile for cat and copy inside attention custom op when piecewise cuda graph is running. For iterations that run under piecewise cuda graph mode, it has a high possibility that they are host-bound cases. Thus, disabling it would improve TTFT.
Summary by CodeRabbit
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.