[TRTLLM-9737][chore] Add rl perf reproduce script and enhance the robustness of Ray tests #9939
Conversation
📝 WalkthroughWalkthroughA new Python script implements a Ray-based distributed inference workflow for TensorRT LLMs, featuring a remote worker class that manages async LLM initialization and token generation, alongside orchestration logic for resource validation, placement group configuration, prompt distribution in round-robin fashion, and async result collection with throughput reporting. Changes
Sequence Diagram(s)sequenceDiagram
participant CLI as CLI Parser
participant Main as main()
participant Setup as setup_rl_llm()
participant Validation as Resource Validator
participant Ray as Ray Environment
participant PG as Placement Group
participant Actors as LLM Instances
participant AsyncCollector as Async Collector
CLI->>Main: parse_arguments()
Main->>Setup: setup_rl_llm(args)
Setup->>Validation: Validate GPU availability
Validation-->>Setup: Validation result
alt Validation Success
Setup->>Ray: Configure Ray environment
Ray-->>Setup: Ray initialized
Setup->>PG: Create STRICT_PACK placement group
PG-->>Setup: Placement group ready
Setup->>Actors: Create num_instances Ray actors (trtllm_instance)
Actors->>Actors: Await actor readiness
Actors-->>Setup: All actors ready
Setup->>Actors: init_llm() on each actor
Actors->>Actors: Initialize AsyncLLM, build SamplingParams
Actors-->>Setup: LLMs initialized
Setup->>AsyncCollector: Distribute prompts round-robin to actors
AsyncCollector->>Actors: generate(prompt)
Actors-->>AsyncCollector: token_ids, logprobs
AsyncCollector-->>Setup: All results collected
Setup->>Setup: Report timing & throughput
end
Setup->>PG: Clean up placement groups
Setup->>Ray: Shutdown Ray
Setup-->>Main: Execution complete
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: 6
🧹 Nitpick comments (6)
examples/ray_orchestrator/rl_perf_repro.py (6)
21-27: Add class and method docstrings.For Python interfaces that may be used outside a file, the coding guidelines prefer docstrings over comments. Consider adding Google-style docstrings to document the class and its parameters.
Example:
@ray.remote class TRTLLMInstance: """Ray remote worker that manages AsyncLLM initialization and generation. Attributes: async_llm_kwargs: Configuration dict for AsyncLLM initialization. sampling_kwargs: Configuration dict for SamplingParams. llm: The AsyncLLM instance (set after init_llm is called). sampling_params: The SamplingParams instance (set after init_llm is called). """ def __init__(self, async_llm_kwargs: dict, sampling_kwargs: dict): """Initialize the TRTLLMInstance with configuration dictionaries. Args: async_llm_kwargs: Configuration parameters for AsyncLLM. sampling_kwargs: Configuration parameters for SamplingParams. """
42-48: Consider making hardcoded AsyncLLM parameters configurable.Several AsyncLLM parameters are hardcoded (
enable_sleep=True,batch_wait_timeout_iters=32,batch_wait_max_tokens_ratio=0.5), which may limit flexibility for different performance reproduction scenarios.
70-70: Simplify logprobs extraction.The single-element slice can be replaced with
next(iter(...))for clarity.Apply this diff:
- log_probs = [list(d.values())[0].logprob for d in outputs.outputs[0].logprobs] + log_probs = [next(iter(d.values())).logprob for d in outputs.outputs[0].logprobs]
74-78: Add function docstring.This is the main orchestration function and should have a comprehensive Google-style docstring documenting its purpose, parameters, and behavior.
84-87: Hardcoded single-node GPU limit.The script enforces a hardcoded limit of 8 GPUs for single-node operation. Consider making this configurable via CLI argument or documenting this limitation more prominently.
268-270: Add main function docstring.Consider adding a docstring to document the entry point's purpose.
def main(): """Entry point for RL performance reproduction script.""" args = parse_arguments() asyncio.run(setup_rl_llm(args))
📜 Review details
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (1)
examples/ray_orchestrator/rl_perf_repro.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:
examples/ray_orchestrator/rl_perf_repro.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:
examples/ray_orchestrator/rl_perf_repro.py
🧠 Learnings (4)
📓 Common learnings
Learnt from: tongyuantongyu
Repo: NVIDIA/TensorRT-LLM PR: 7520
File: tensorrt_llm/_torch/pyexecutor/resource_manager.py:605-613
Timestamp: 2025-09-24T03:31:28.908Z
Learning: In TensorRT-LLM Ray orchestrator mode, ProcessGroups are initialized with both Gloo and NCCL backends (e.g., "cuda:nccl,cpu:gloo"), allowing PyTorch distributed to automatically route CPU tensors through Gloo and GPU tensors through NCCL. This eliminates the need for manual device placement when performing allreduce operations on base types.
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.
Learnt from: ixlmar
Repo: NVIDIA/TensorRT-LLM PR: 7294
File: tensorrt_llm/_torch/modules/rms_norm.py:17-17
Timestamp: 2025-08-27T14:23:55.566Z
Learning: The TensorRT-LLM project requires Python 3.10+ as evidenced by the use of TypeAlias from typing module, match/case statements, and union type | syntax throughout the codebase, despite some documentation still mentioning Python 3.8+.
Learnt from: achartier
Repo: NVIDIA/TensorRT-LLM PR: 6763
File: tests/integration/defs/triton_server/conftest.py:16-22
Timestamp: 2025-08-11T20:09:24.389Z
Learning: In the TensorRT-LLM test infrastructure, the team prefers simple, direct solutions (like hard-coding directory traversal counts) over more complex but robust approaches when dealing with stable directory structures. They accept the maintenance cost of updating tests if the layout changes.
Learnt from: tongyuantongyu
Repo: NVIDIA/TensorRT-LLM PR: 7763
File: cpp/tensorrt_llm/CMakeLists.txt:297-301
Timestamp: 2025-09-16T09:30:09.716Z
Learning: In the TensorRT-LLM project, NCCL libraries are loaded earlier by PyTorch libraries or the bindings library, so the main shared library doesn't need NCCL paths in its RPATH - the libraries will already be available in the process address space when needed.
Learnt from: yibinl-nvidia
Repo: NVIDIA/TensorRT-LLM PR: 6506
File: examples/models/core/mixtral/requirements.txt:3-3
Timestamp: 2025-08-01T15:14:45.673Z
Learning: In TensorRT-LLM, examples directory can have different dependency versions than the root requirements.txt file. Version conflicts between root and examples dependencies are acceptable because examples are designed to be standalone and self-contained.
Learnt from: nv-lschneider
Repo: NVIDIA/TensorRT-LLM PR: 7910
File: cpp/tensorrt_llm/kernels/nccl_device/config.cu:42-49
Timestamp: 2025-09-23T14:58:05.372Z
Learning: In TensorRT-LLM NCCL device kernels (cpp/tensorrt_llm/kernels/nccl_device/), the token partitioning intentionally uses ceil-like distribution (same token_per_rank for all ranks) to ensure all ranks launch the same number of blocks. This is required for optimal NCCL device API barrier performance, even though it may launch extra blocks for non-existent tokens on later ranks. Runtime bounds checking in the kernel (blockID validation) handles the overshoot cases.
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.
Learnt from: dcampora
Repo: NVIDIA/TensorRT-LLM PR: 6867
File: tensorrt_llm/_torch/pyexecutor/sampler.py:67-72
Timestamp: 2025-08-13T16:20:37.987Z
Learning: In TensorRT-LLM sampler code, performance is prioritized over additional validation checks. The beam_width helper method intentionally returns the first request's beam_width without validating consistency across all requests to avoid performance overhead from iterating through the entire batch.
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.
📚 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:
examples/ray_orchestrator/rl_perf_repro.py
📚 Learning: 2025-09-24T03:31:28.908Z
Learnt from: tongyuantongyu
Repo: NVIDIA/TensorRT-LLM PR: 7520
File: tensorrt_llm/_torch/pyexecutor/resource_manager.py:605-613
Timestamp: 2025-09-24T03:31:28.908Z
Learning: In TensorRT-LLM Ray orchestrator mode, ProcessGroups are initialized with both Gloo and NCCL backends (e.g., "cuda:nccl,cpu:gloo"), allowing PyTorch distributed to automatically route CPU tensors through Gloo and GPU tensors through NCCL. This eliminates the need for manual device placement when performing allreduce operations on base types.
Applied to files:
examples/ray_orchestrator/rl_perf_repro.py
📚 Learning: 2025-11-27T09:23:18.742Z
Learnt from: fredricz-20070104
Repo: NVIDIA/TensorRT-LLM PR: 9511
File: tests/integration/defs/examples/serve/test_serve.py:136-186
Timestamp: 2025-11-27T09:23:18.742Z
Learning: In TensorRT-LLM testing, when adding test cases based on RCCA commands, the command format should be copied exactly as it appears in the RCCA case, even if it differs from existing tests. For example, some RCCA commands for trtllm-serve may omit the "serve" subcommand while others include it.
Applied to files:
examples/ray_orchestrator/rl_perf_repro.py
🧬 Code graph analysis (1)
examples/ray_orchestrator/rl_perf_repro.py (3)
tensorrt_llm/_torch/async_llm.py (2)
AsyncLLM(7-106)setup_async(40-46)tensorrt_llm/llmapi/llm_args.py (1)
CudaGraphConfig(107-164)tensorrt_llm/llmapi/llm.py (1)
prompt(87-88)
🪛 Ruff (0.14.8)
examples/ray_orchestrator/rl_perf_repro.py
70-70: Prefer next(iter(d.values())) over single element slice
Replace with next(iter(d.values()))
(RUF015)
85-87: Avoid specifying long messages outside the exception class
(TRY003)
89-91: Avoid specifying long messages outside the exception class
(TRY003)
185-185: Local variable results is assigned to but never used
Remove assignment to unused variable results
(F841)
⏰ 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 (4)
examples/ray_orchestrator/rl_perf_repro.py (4)
64-64: Verify Python version compatibility for type annotation syntax.The type annotation
list[int]uses Python 3.9+ syntax, but the coding guidelines state "The code developed for TensorRT-LLM should conform to Python 3.8+". For Python 3.8 compatibility, useList[int]from thetypingmodule instead.However, based on learnings, TensorRT-LLM may actually require Python 3.10+ in practice. Please verify the project's actual Python version requirement.
If Python 3.8 compatibility is required, apply this diff:
+from typing import List + ... - async def generate(self, prompt: list[int]): + async def generate(self, prompt: List[int]):
135-135: Clarify conditional max_batch_size logic.The logic
"max_batch_size": 0 if args.batch_sizes else args.max_batch_sizesetsmax_batch_sizeto 0 when custombatch_sizesare provided. Please verify this is the intended behavior forCudaGraphConfig, as setting it to 0 seems unusual.Based on the relevant code snippets,
CudaGraphConfighas avalidate_cuda_graph_max_batch_sizemethod that "ensures max_batch_size is non-negative." Setting it to 0 might bypass CUDA graph generation or use a default behavior. Please confirm this aligns with the intended configuration.
155-155: Hardcoded end_id may cause issues.The
end_idis hardcoded to-1, which typically means "no end token". Verify this is appropriate for all models, as some models may require a valid end token ID for proper generation termination.
127-127: Worker extension class exists and is appropriate.The
WorkerExtensionclass is properly defined intensorrt_llm/llmapi/rlhf_utils.pyand is designed specifically for this purpose with methods for updating weights and checking weight updates—both essential for RLHF workflows.
shuyixiong
force-pushed
the
user/shuyix/rl_perf_repro
branch
from
98e600f to
fdd2dcb
Compare
shuyixiong
force-pushed
the
user/shuyix/rl_perf_repro
branch
from
575877c to
53e8c12
Compare
|
/bot run --stage-list "DGX_H100-2_GPUs-PyTorch-Ray-1,DGX_H100-4_GPUs-PyTorch-Ray-1" |
|
PR_Github #28764 [ run ] triggered by Bot. Commit: |
|
PR_Github #28764 [ run ] completed with state
|
|
/bot run --stage-list "DGX_H100-2_GPUs-PyTorch-Ray-1,DGX_H100-4_GPUs-PyTorch-Ray-1" |
|
PR_Github #28774 [ run ] triggered by Bot. Commit: |
|
PR_Github #28774 [ run ] completed with state
|
|
/bot run --stage-list "DGX_H100-2_GPUs-PyTorch-Ray-1,DGX_H100-4_GPUs-PyTorch-Ray-1" |
|
PR_Github #28795 [ run ] triggered by Bot. Commit: |
tburt-nv
left a comment
tburt-nv
left a comment
There was a problem hiding this comment.
Approving the pipeline change, although it looks like the tests are failing and may need more work.
|
PR_Github #28795 [ run ] completed with state
|
|
/bot run --stage-list "DGX_H100-2_GPUs-PyTorch-Ray-1,DGX_H100-4_GPUs-PyTorch-Ray-1" |
|
/bot run --stage-list "DGX_H100-2_GPUs-PyTorch-Ray-1,DGX_H100-4_GPUs-PyTorch-Ray-1" --disable-reuse-test |
|
PR_Github #29517 [ run ] triggered by Bot. Commit: |
|
PR_Github #29518 [ run ] triggered by Bot. Commit: |
|
PR_Github #29518 [ run ] completed with state |
Signed-off-by: Shuyi Xiong <[email protected]>
|
/bot run --stage-list "DGX_H100-2_GPUs-PyTorch-Ray-1,DGX_H100-4_GPUs-PyTorch-Ray-1" |
|
PR_Github #29543 [ run ] triggered by Bot. Commit: |
|
PR_Github #29543 [ run ] completed with state
|
Signed-off-by: Shuyi Xiong <[email protected]>
|
/bot run --stage-list "DGX_H100-2_GPUs-PyTorch-Ray-1,DGX_H100-4_GPUs-PyTorch-Ray-1" |
|
PR_Github #29570 [ run ] triggered by Bot. Commit: |
Signed-off-by: Shuyi Xiong <[email protected]>
|
PR_Github #29570 [ run ] completed with state |
shuyixiong
changed the title
|
/bot run --disable-fail-fast |
|
PR_Github #29605 [ run ] triggered by Bot. Commit: |
|
PR_Github #29605 [ run ] completed with state
|
|
/bot run --disable-fail-fast |
|
PR_Github #29721 [ run ] triggered by Bot. Commit: |
|
PR_Github #29721 [ run ] completed with state |
…ustness of Ray tests (NVIDIA#9939) Signed-off-by: Shuyi Xiong <[email protected]>
…ustness of Ray tests (NVIDIA#9939) Signed-off-by: Shuyi Xiong <[email protected]> Signed-off-by: Zihua Wu <[email protected]>
…ustness of Ray tests (NVIDIA#9939) Signed-off-by: Shuyi Xiong <[email protected]>
6 participants
Summary by CodeRabbit
Examples
Tests
RAY_ADDRESSto ensure connection to the specified cluster, thereby eliminating interference between Ray stages caused by shared Ray status.CI
✏️ 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.