[TRTLLM-6960][fix] replace flasky scaled_mm test with more stable config #7089
Conversation
📝 WalkthroughWalkthroughParameterization of a unit test changed (m: 12 → 8), an SM90-specific skip inside Changes
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes Possibly related PRs
Suggested reviewers
Tip 🔌 Remote MCP (Model Context Protocol) integration is now available!Pro plan users can now connect to remote MCP servers from the Integrations page. Connect with popular remote MCPs such as Notion and Linear to add more context to your reviews and chats. 📜 Recent review detailsConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro 💡 Knowledge Base configuration:
You can enable these sources in your CodeRabbit configuration. 📒 Files selected for processing (1)
🚧 Files skipped from review as they are similar to previous 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 --stage-list "H100_PCIe-PyTorch-2" |
![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
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
tests/unittest/_torch/thop/test_scaled_mm.py (1)
61-72: Restore CUBLASLT_WORKSPACE_SIZE safely and set the correct “32MB” value.Two issues:
- The value f"{32*1024}" is 32 KiB, not 32 MiB as the comment states. This undermines the intent to align with a 32MB workspace and can change algorithm selection.
- The env var is restored unconditionally only on the happy path. If an exception/assert fires before Line 71, the env var leaks into subsequent tests.
Wrap the workspace override in try/finally, restore the previous state (including unsetting if it wasn’t defined), and set the correct bytes for 32MB.
Apply this diff:
- old_env = os.environ.get("CUBLASLT_WORKSPACE_SIZE", "") - os.environ["CUBLASLT_WORKSPACE_SIZE"] = f"{32*1024}" - ref = torch._scaled_mm( - x, - w.t(), - out_dtype=output_dtype, - scale_a=scale_x, - scale_b=scale_w, - use_fast_accum=True, - ) - os.environ["CUBLASLT_WORKSPACE_SIZE"] = old_env + had_cublaslt_ws = "CUBLASLT_WORKSPACE_SIZE" in os.environ + old_cublaslt_ws = os.environ.get("CUBLASLT_WORKSPACE_SIZE") + # 32 MiB, in bytes + os.environ["CUBLASLT_WORKSPACE_SIZE"] = str(32 * 1024 * 1024) + try: + ref = torch._scaled_mm( + x, + w.t(), + out_dtype=output_dtype, + scale_a=scale_x, + scale_b=scale_w, + use_fast_accum=True, + ) + finally: + if had_cublaslt_ws: + os.environ["CUBLASLT_WORKSPACE_SIZE"] = old_cublaslt_ws + else: + os.environ.pop("CUBLASLT_WORKSPACE_SIZE", None)
🧹 Nitpick comments (5)
tests/unittest/_torch/thop/test_scaled_mm.py (5)
72-75: Tightened atol=0.01: better signal, but consider dtype-aware tolerances to avoid future flakes.A flat atol=1e-2 with rtol=1e-2 may be tight for some FP8 paths depending on toolkit/driver micro-versions. Using per-output-dtype tolerances tends to be more resilient while remaining strict.
Apply this diff:
- np.testing.assert_allclose(ref.float().cpu(), - output.float().cpu(), - atol=0.01, - rtol=0.01) + tol_map = { + torch.float32: (1e-2, 1e-2), + torch.float16: (2e-2, 1e-2), + torch.bfloat16: (5e-2, 2e-2), + } + atol, rtol = tol_map[output_dtype] + np.testing.assert_allclose( + ref.float().cpu(), + output.float().cpu(), + atol=atol, + rtol=rtol, + )
24-27: Skip condition is fine; add an operator-availability guard to turn hard errors into skips.If the custom op isn’t registered (e.g., build variants without trtllm ops), the test will error out instead of skip. Guard early to improve CI ergonomics on non-standard builds.
Apply this diff near the top of the test (right after the decorators or at the start of the function):
@pytest.mark.parametrize( "output_dtype", [torch.float16, torch.float32, torch.bfloat16], ) def test_fp8_scaled_mm(output_dtype, m, k_n): + if not hasattr(torch.ops, "trtllm") or not hasattr(torch.ops.trtllm, "cublas_scaled_mm"): + pytest.skip("trtllm.cublas_scaled_mm op not available in this build")
57-60: Fix comment typos and proper capitalization.Minor nits: “cause cause” duplication and “pytorch” capitalization.
Apply this diff:
- # If anywhere else calls torch's cublas op, the static workspace size will - # be fixed to 1MB. If not aligned, will cause cause pytorch not using splitK + # If anywhere else calls torch's cuBLAS op, the static workspace size will + # be fixed to 1MB. If not aligned, it may cause PyTorch not to use splitK
15-17: Use warnings.warn with category instead of passing an exception instance.Passing a RuntimeWarning instance to warn() doesn’t set the category. Prefer warnings.warn(message, RuntimeWarning) and import warnings accordingly.
Apply this diff:
-import os -from warnings import warn +import os +import warnings @@ - except Exception as e: - warn(RuntimeWarning("cutlass result is not correct: " + repr(e))) + except Exception as e: + warnings.warn("cutlass result is not correct: " + repr(e), RuntimeWarning)Also applies to: 92-93
96-98: Remove ad-hoc main invocation or update it.Directly invoking the test with m=12 contradicts the new parametrization and bypasses pytest decorators/skips. This is confusing and can fail outside SM90.
Apply this diff to remove it:
-if __name__ == '__main__': - test_fp8_scaled_mm(torch.float16, 12, (8192, 10240)) +# Intentionally no __main__ entry; run via pytest to honor decorators.
📜 Review details
Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro
💡 Knowledge Base configuration:
- MCP integration is disabled by default for public repositories
- Jira integration is disabled by default for public repositories
- Linear integration is disabled by default for public repositories
You can enable these sources in your CodeRabbit configuration.
📒 Files selected for processing (1)
tests/unittest/_torch/thop/test_scaled_mm.py(2 hunks)
🧰 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:
tests/unittest/_torch/thop/test_scaled_mm.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:
tests/unittest/_torch/thop/test_scaled_mm.py
⏰ 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)
tests/unittest/_torch/thop/test_scaled_mm.py (1)
33-35: Param m: switch 12 → 8 improves stability; good call.Reducing to 8 is a pragmatic way to avoid shape-driven algorithm variance. Given the remaining shapes [2048, 228], coverage looks reasonable.
If you’ve seen flakiness tied to specific m mod factors (e.g., splitK thresholds), consider confirming that these three shapes still exercise the same code paths on your CI SM90 fleet (different CUDA/toolkit versions).
|
PR_Github #15897 [ run ] triggered by Bot |
|
PR_Github #15897 [ run ] completed with state |
dc3671
force-pushed
the
fix-scaled-mm-test
branch
from
19e6acd to
8b0f8ab
Compare
|
/bot run --stage-list "H100_PCIe-PyTorch-2" |
|
PR_Github #15996 [ run ] triggered by Bot |
|
PR_Github #15996 [ run ] completed with state |
dc3671
force-pushed
the
fix-scaled-mm-test
branch
from
8b0f8ab to
4cb802e
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #16052 [ run ] triggered by Bot |
|
PR_Github #16052 [ run ] completed with state |
|
/bot run --disable-fail-fast |
|
PR_Github #16119 [ run ] triggered by Bot |
dc3671
force-pushed
the
fix-scaled-mm-test
branch
from
4cb802e to
f80d659
Compare
|
/bot run --disable-fail-fast |
|
PR_Github #16119 [ run ] completed with state |
|
PR_Github #16160 [ run ] triggered by Bot |
|
PR_Github #16160 [ run ] completed with state |
dc3671
force-pushed
the
fix-scaled-mm-test
branch
from
f80d659 to
8609e22
Compare
|
/bot run |
|
PR_Github #16343 [ run ] triggered by Bot |
|
PR_Github #16343 [ run ] completed with state |
Signed-off-by: Zhenhuan Chen <chenzhh3671@gmail.com>
dc3671
force-pushed
the
fix-scaled-mm-test
branch
from
8609e22 to
673dd4e
Compare
|
/bot run |
|
PR_Github #16400 [ run ] triggered by Bot |
|
PR_Github #16400 [ run ] completed with state |
|
@hlu1 What do you think about this modification to the test? It can pass serveral times and will not trigger different algos problem again. |
…fig (NVIDIA#7089) Signed-off-by: Zhenhuan Chen <chenzhh3671@gmail.com>
3 participants
Summary by CodeRabbit
Description
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.
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.