[https://nvbugs/5445466][fix] fix deepseek r1 hang by not enabling mnnvl by default #6860
Conversation
Signed-off-by: Pengbo Wang <[email protected]>
📝 WalkthroughWalkthroughNarrowed MNNVL AllReduce initialization to only the explicit MNNVL strategy and left forward-path fallback logic unchanged. Removed two DeepSeekR1 latency waivers from tests/integration/test_lists/waives.txt. Changes
Sequence Diagram(s)sequenceDiagram
participant Caller
participant AllReduceOp
participant MNNVL
participant DefaultAR
Caller->>AllReduceOp: forward(tensor)
alt MNNVL instance available
AllReduceOp->>MNNVL: allreduce(tensor)
alt MNNVL returns result
MNNVL-->>AllReduceOp: result
AllReduceOp-->>Caller: result
else MNNVL returns None/error
AllReduceOp->>DefaultAR: allreduce(tensor) + set strategy=AUTO if was MNNVL
DefaultAR-->>AllReduceOp: result
AllReduceOp-->>Caller: result
end
else No MNNVL instance (AUTO init no longer creates)
AllReduceOp->>DefaultAR: allreduce(tensor)
DefaultAR-->>AllReduceOp: result
AllReduceOp-->>Caller: result
end
Estimated code review effort🎯 2 (Simple) | ⏱️ ~8 minutes Possibly related PRs
Suggested reviewers
📜 Recent review detailsConfiguration used: .coderabbit.yaml 📒 Files selected for processing (1)
💤 Files with no reviewable changes (1)
⏰ 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)
✨ Finishing Touches🧪 Generate unit tests
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. 🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
SupportNeed help? Create a ticket on our support page for assistance with any issues or questions. CodeRabbit Commands (Invoked using PR/Issue comments)Type Other keywords and placeholders
Status, Documentation and Community
|
|
/bot run |
![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 (3)
tensorrt_llm/_torch/distributed/ops.py (3)
458-476: Log a clear reason when dtype is None and avoid noisy multi-rank logs.If users request MNNVL but forget to pass dtype, the current debug message blames the is_mnnvl check. Emit an explicit warning for dtype=None and only log once (rank 0) to reduce noise.
Apply this diff:
@@ - if self.strategy == AllReduceStrategy.MNNVL: - if MNNVLAllReduce.is_mnnvl(self.mapping, dtype): - try: - self.mnnvl_allreduce = MNNVLAllReduce( - self.mapping, dtype) if dtype else None - if self.mnnvl_allreduce: - logger.debug(f"MNNVLAllReduce is enabled") - else: - logger.debug(f"MNNVLAllReduce is disabled") - except Exception as e: - logger.debug( - f"MNNVL AllReduce can't be enabled due to {e}.") - self.mnnvl_allreduce = None - else: - logger.debug( - f"MNNVLAllReduce can't be enabled due to failing the is_mnnvl check." - ) - self.mnnvl_allreduce = None + if self.strategy == AllReduceStrategy.MNNVL: + if dtype is None: + if getattr(self.mapping, "tp_rank", 0) == 0: + logger.warning("MNNVL strategy requested but dtype is None; disabling MNNVL.") + self.mnnvl_allreduce = None + elif MNNVLAllReduce.is_mnnvl(self.mapping, dtype): + try: + self.mnnvl_allreduce = MNNVLAllReduce(self.mapping, dtype) + if getattr(self.mapping, "tp_rank", 0) == 0: + logger.debug("MNNVLAllReduce is enabled") + except Exception as e: + if getattr(self.mapping, "tp_rank", 0) == 0: + logger.debug(f"MNNVL AllReduce can't be enabled due to {e}.") + self.mnnvl_allreduce = None + else: + if getattr(self.mapping, "tp_rank", 0) == 0: + logger.debug("MNNVLAllReduce can't be enabled due to failing the is_mnnvl check.") + self.mnnvl_allreduce = None
409-444: Docstring omission: include MNNVL strategy and its new init behavior.The constructor docs enumerate strategies but omit MNNVL. Add it and clarify it’s only instantiated when explicitly selected.
Apply this doc update:
@@ - - LOWPRECISION: AllReduce quantizes data to lower precision for transmission. + - LOWPRECISION: AllReduce quantizes data to lower precision for transmission. Should only be used on topologies with PCIe switches and without NVLink. This strategy may result in some precision loss but can improve performance on specific hardware configurations. + + - MNNVL: Use Multi-Node NVLink-optimized AllReduce for supported platforms and dtypes. + Note: MNNVL kernels are only initialized when strategy is explicitly set to MNNVL. + AUTO will not instantiate or attempt MNNVL during initialization.
1-1: Missing NVIDIA copyright header.Per repository guidelines, prepend the current year NVIDIA copyright header.
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (2)
tensorrt_llm/_torch/distributed/ops.py(1 hunks)tests/integration/test_lists/waives.txt(0 hunks)
💤 Files with no reviewable changes (1)
- tests/integration/test_lists/waives.txt
🧰 Additional context used
📓 Path-based instructions (2)
**/*.py
📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)
**/*.py: Python code must target Python 3.8+
Python indentation: 4 spaces, no tabs
Maintain module namespace in imports (from package.subpackage import foo; then use foo.SomeClass())
Python file names use snake_case
Python class names use PascalCase
Python functions/methods and local variables use snake_case; variables starting with a number get k_ prefix (e.g., k_99th_percentile)
Global variables use G_ prefixed UPPER_SNAKE_CASE (e.g., G_MY_GLOBAL)
Constants use UPPER_SNAKE_CASE in Python
Avoid shadowing variables from outer scopes in Python
Initialize all externally visible members of a Python class in init
Prefer docstrings for interfaces used outside a file; comments for local code
Use Google-style docstrings for classes and functions (Sphinx-parsable)
Document attributes/variables inline with short docstrings
Avoid reflection when simple alternatives exist (e.g., prefer explicit parameters over dict(**locals()))
In try/except, catch the narrowest exceptions possible
For duck-typing with try/except, keep try body minimal and put logic in else
Files:
tensorrt_llm/_torch/distributed/ops.py
**/*.{cpp,cxx,cc,cu,h,hpp,hxx,hh,cuh,py}
📄 CodeRabbit Inference Engine (CODING_GUIDELINES.md)
Prepend NVIDIA copyright header (current year) to all source files
Files:
tensorrt_llm/_torch/distributed/ops.py
🔇 Additional comments (2)
tensorrt_llm/_torch/distributed/ops.py (2)
458-476: Explicitly gating MNNVL init behind strategy==MNNVL is correct and aligns with the PR goal.This prevents accidental MNNVL buffer initialization in AUTO mode and directly addresses the DeepSeek R1 hang.
526-528: Fallback guard for MNNVL is correctly implementedThe check in tensorrt_llm/_torch/distributed/ops.py (lines 526–528) remaps MNNVL to AUTO before invoking torch.ops.trtllm.allreduce, so we never pass an unsupported strategy to the custom op. A code-wide search confirms this is the only path where MNNVL is converted to AUTO for the torch operator. No further action needed.
|
PR_Github #15106 [ run ] triggered by Bot |
|
PR_Github #15106 [ run ] completed with state |
|
/bot run --add-multi-gpu-test --disable-fail-fast |
|
PR_Github #15120 [ run ] triggered by Bot |
|
PR_Github #15120 [ run ] completed with state |
|
/bot run --add-multi-gpu-test --disable-fail-fast |
1 similar comment
|
/bot run --add-multi-gpu-test --disable-fail-fast |
|
PR_Github #15197 [ run ] triggered by Bot |
|
/bot kill |
|
PR_Github #15276 [ kill ] triggered by Bot |
|
PR_Github #15276 [ kill ] completed with state |
|
/bot run --add-multi-gpu-test --disable-fail-fast |
|
PR_Github #15278 [ run ] triggered by Bot |
|
The A10-PyTorch-1/disaggregated/test_disaggregated.py::test_disaggregated_diff_max_tokens[TinyLlama-1.1B-Chat-v1.0] is unstable, already failed in the post merge https://prod.blsm.nvidia.com/sw-tensorrt-top-1/job/LLM/job/main/job/L0_PostMerge/2232/ Not related. |
|
I am forcing merge it. Since all the failed cases are already failed in post merge, and not related to this one. |
|
PR_Github #15278 [ run ] completed with state |
|
note that this PR has passed single and multi gpu run of https://nv/trt-llm-cicd/job/helpers/job/PR_Github/15197/ despite not being reported due to ci issue. |
…nvl by default (NVIDIA#6860) Signed-off-by: Pengbo Wang <[email protected]> Co-authored-by: Tao Li @ NVIDIA <[email protected]>
…nvl by default (NVIDIA#6860) Signed-off-by: Pengbo Wang <[email protected]> Co-authored-by: Tao Li @ NVIDIA <[email protected]> Signed-off-by: Wangshanshan <[email protected]>
…nvl by default (NVIDIA#6860) Signed-off-by: Pengbo Wang <[email protected]> Co-authored-by: Tao Li @ NVIDIA <[email protected]> Signed-off-by: Wangshanshan <[email protected]>
…nvl by default (NVIDIA#6860) Signed-off-by: Pengbo Wang <[email protected]> Co-authored-by: Tao Li @ NVIDIA <[email protected]>
…nvl by default (NVIDIA#6860) Signed-off-by: Pengbo Wang <[email protected]> Co-authored-by: Tao Li @ NVIDIA <[email protected]> Signed-off-by: Wangshanshan <[email protected]>
…nvl by default (NVIDIA#6860) Signed-off-by: Pengbo Wang <[email protected]> Co-authored-by: Tao Li @ NVIDIA <[email protected]>
…nvl by default (NVIDIA#6860) Signed-off-by: Pengbo Wang <[email protected]> Co-authored-by: Tao Li @ NVIDIA <[email protected]> Signed-off-by: Wangshanshan <[email protected]>
Summary by CodeRabbit
New Features
Bug Fixes
Tests
Description
Fix Deepseek R1 hang issue. The hang was caused by MNNVL Buffer initialization process.
Test Coverage
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.