[https://nvbugs/5451280][fix] Reduce memory fraction problem by warmu… #7999
Conversation
|
/bot run |
|
PR_Github #19954 [ run ] triggered by Bot |
📝 WalkthroughWalkthroughIntroduces a generalized warmup routine with a new get_warmup_request parameter to choose between legacy and alternate context computations. Adds general_warmup utility and integrates it into PyTorch and piecewise CUDA-graph warmup flows, including a “most-requests” pass and optional reverse ordering. Updates control flow to use these routines and synchronizations. Changes
Sequence Diagram(s)sequenceDiagram
autonumber
participant User as Caller
participant Engine as ModelEngine
participant Warm as general_warmup()
participant Req as get_warmup_request()
participant Model as forward()
participant CUDA as cuda.synchronize()
Note over Engine: _torch_compile_enabled path
User->>Engine: warmup()
Engine->>Warm: general_warmup(reverse=false)
Warm->>Req: (1,1) via least_requrests=true
Req-->>Warm: batch
Warm->>Model: forward(batch)
Warm->>CUDA: synchronize
Warm->>Req: (batch_size,batch_size)
Req-->>Warm: batch
Warm->>Model: forward(batch)
Warm->>CUDA: synchronize
Warm->>Req: (2,0)
Req-->>Warm: batch
Warm->>Model: forward(batch)
Warm->>CUDA: synchronize
Warm->>Req: (curr_max_num_tokens,0)
Req-->>Warm: batch
Warm->>Model: forward(batch)
Warm->>CUDA: synchronize
Warm-->>Engine: done
sequenceDiagram
autonumber
participant User as Caller
participant Engine as ModelEngine
participant Req as get_warmup_request()
participant Warm as general_warmup()
participant Model as forward()
participant CUDA as cuda.synchronize()
Note over Engine: piecewise_cuda_graph path
User->>Engine: warmup()
Engine->>Req: (num_tokens, 0, least_requrests=false) %% "most-requests" batch
Req-->>Engine: batch_max_reqs
Engine->>Model: forward(batch_max_reqs)
Engine->>CUDA: synchronize
Note over Engine,Warm: Reverse pass to allocate large→small
Engine->>Warm: general_warmup(reverse=true)
Warm->>Req: (curr_max_num_tokens,0) ...
loop predefined set (reversed)
Req-->>Warm: batch
Warm->>Model: forward(batch)
Warm->>CUDA: synchronize
end
Note over Engine: Optional additional pass repeated per code
Engine->>Model: forward(batch_max_reqs)
Engine->>CUDA: synchronize
Engine-->>User: warmup complete
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
Tip 👮 Agentic pre-merge checks are now available in preview!Pro plan users can now enable pre-merge checks in their settings to enforce checklists before merging PRs.
Please see the documentation for more information. Example: reviews:
pre_merge_checks:
custom_checks:
- name: "Undocumented Breaking Changes"
mode: "warning"
instructions: |
Pass/fail criteria: All breaking changes to public APIs, CLI flags, environment variables, configuration keys, database schemas, or HTTP/GraphQL endpoints must be documented in the "Breaking Change" section of the PR description and in CHANGELOG.md. Exclude purely internal or private changes (e.g., code not exported from package entry points or explicitly marked as internal).Please share your feedback with us on this Discord post. 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: 2
🧹 Nitpick comments (3)
tensorrt_llm/_torch/pyexecutor/model_engine.py (3)
769-794: Make warmup ordering deterministic and reduce log verbosity.
- Using a set results in non-deterministic iteration order; this can lead to inconsistent warmup patterns and memory behavior across runs.
- Prefer a list with fixed order; sort only when
reverse=True.- Consider downgrading logs to debug to avoid noisy INFO logs during warmup.
Apply this diff:
- def general_warmup(reverse: bool = False): - warmup_requests = set([ - (1, 1), # Specialize for 1 token. - (self.batch_size, - self.batch_size), # max_batch_size, pure generation - (2, 0), # Non-one, pure context - (curr_max_num_tokens, 0), # max_num_tokens, pure context - ]) - if reverse: - warmup_requests = sorted(list(warmup_requests), reverse=reverse) + def general_warmup(reverse: bool = False): + warmup_requests = [ + (1, 1), # Specialize for 1 token. + (self.batch_size, self.batch_size), # max_batch_size, pure generation + (2, 0), # Non-one, pure context + (curr_max_num_tokens, 0), # max_num_tokens, pure context + ] + if reverse: + warmup_requests = sorted(warmup_requests, key=lambda x: (x[0], x[1]), reverse=True) @@ - logger.info( + logger.debug( f"Run warmup with {warmup_num_tokens} tokens, include {warmup_num_gen_tokens} generation tokens" )
807-807: Good consolidation.Replacing the manual loop with
general_warmup()improves reuse and readability. Note: thewarmup_requestsvariable defined above (Lines 797–804) is now unused; consider removing it in a follow-up cleanup.
894-914: Finalize “most-requests” pass; fix comment typos and ensure clarity.
- Typo: “memory faction” → “memory fragmentation”.
- Clarify the comment text.
Apply this diff:
- # When using piecewise cuda graph, the logits may suffer severe memory faction problem. - # When the num of requests is growing, the block allocated by torch cannot be reused. - # So after piecewise cuda graph capture, a request with most requests is triggered to makes - # sure that a large enough block is allocated and can be correctly reused. + # When using piecewise CUDA graph, logits allocations may suffer severe memory fragmentation. + # As the number of requests grows, blocks allocated by torch may not be reused efficiently. + # After piecewise CUDA graph capture, trigger a batch with the most requests to ensure + # a sufficiently large block is allocated and can be correctly reused. @@ - with release_batch(batch) as batch: + with release_batch(batch) as batch: logger.info( f"Run piecewise CUDA graph warmup for num tokens={num_tokens} with most requests" )
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (1)
tensorrt_llm/_torch/pyexecutor/model_engine.py(5 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/pyexecutor/model_engine.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/pyexecutor/model_engine.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/pyexecutor/model_engine.py
🧠 Learnings (1)
📚 Learning: 2025-08-19T12:45:11.997Z
Learnt from: amitz-nv
PR: NVIDIA/TensorRT-LLM#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/pyexecutor/model_engine.py
🧬 Code graph analysis (1)
tensorrt_llm/_torch/pyexecutor/model_engine.py (3)
tensorrt_llm/_torch/attention_backend/trtllm.py (2)
max_seq_len(571-581)max_seq_len(584-588)tensorrt_llm/_torch/pyexecutor/scheduler.py (1)
batch_size(35-36)tensorrt_llm/logger.py (1)
info(138-139)
⏰ 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
|
PR_Github #19954 [ run ] completed with state |
liji-nv
force-pushed
the
dev-liji-compile-mem-opt
branch
2 times, most recently
from
4d5ebb7 to
e1d60cc
Compare
|
/bot run |
|
PR_Github #20082 [ run ] triggered by Bot |
liji-nv
force-pushed
the
dev-liji-compile-mem-opt
branch
from
e1d60cc to
016c0bd
Compare
liji-nv
requested review from
QiJune,
nv-guomingz,
tburt-nv and
yuanjingx87
|
/bot run |
|
PR_Github #20087 [ run ] triggered by Bot |
|
PR_Github #20082 [ run ] completed with state |
|
PR_Github #20087 [ run ] completed with state |
|
PR_Github #20230 [ run ] triggered by Bot |
liji-nv
force-pushed
the
dev-liji-compile-mem-opt
branch
from
a745b22 to
bad7eeb
Compare
|
PR_Github #20230 [ run ] completed with state |
…p with large request Signed-off-by: Jin Li <[email protected]>
liji-nv
force-pushed
the
dev-liji-compile-mem-opt
branch
from
bad7eeb to
d881441
Compare
|
/bot run |
|
PR_Github #20263 [ run ] triggered by Bot |
|
PR_Github #20263 [ run ] completed with state |
|
/bot run |
|
PR_Github #20321 [ run ] triggered by Bot |
|
PR_Github #20321 [ run ] completed with state |
|
/bot run |
|
PR_Github #20354 [ run ] triggered by Bot |
|
PR_Github #20354 [ run ] completed with state |
|
/bot run |
|
PR_Github #20497 [ run ] triggered by Bot |
|
PR_Github #20497 [ run ] completed with state |
|
/bot run |
|
PR_Github #20621 [ run ] triggered by Bot |
|
PR_Github #20621 [ run ] completed with state |
NVIDIA#7999) Signed-off-by: Jin Li <[email protected]>
NVIDIA#7999) Signed-off-by: Jin Li <[email protected]>
NVIDIA#7999) Signed-off-by: Jin Li <[email protected]> Signed-off-by: Mike Iovine <[email protected]>
NVIDIA#7999) Signed-off-by: Jin Li <[email protected]>
NVIDIA#7999) Signed-off-by: Jin Li <[email protected]>
NVIDIA#7999) Signed-off-by: Jin Li <[email protected]> Signed-off-by: Mike Iovine <[email protected]>
…p with large request
Summary by CodeRabbit
Performance Improvements
Reliability
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.