refine(repro): enable deterministic algorithms - #1307
Merged
Merged
Conversation
- seed_all now calls torch.use_deterministic_algorithms(True, warn_only=True) after the cuDNN flags, so every op with a deterministic kernel uses it and the rest warn at execution time instead of raising - guard the call with try/except RuntimeError, logging via get_logger so a failure to enable determinism degrades to a warning rather than propagating out of seed_all - import get_logger in reproducibility - test_train_convergence_segmentation: raise epochs 5 -> 6 and update the docstring epoch references, widening the convergence margin above GPU run-to-run noise --- Co-authored-by: claude[bot] <209825114+claude[bot]@users.noreply.github.com> Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Contributor
There was a problem hiding this comment.
Pull request overview
This PR targets improved experiment reproducibility by enabling PyTorch’s deterministic algorithms mode (best-effort, warning on unsupported ops) and adjusts the synthetic segmentation training convergence benchmark to run for one additional epoch to stabilize results.
Changes:
- Update
seed_all()to calltorch.use_deterministic_algorithms(True, warn_only=True)and warn (instead of raising) if determinism cannot be enabled. - Update the synthetic segmentation convergence benchmark to train for 6 epochs (and update the corresponding docstring assertions).
Reviewed changes
Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.
| File | Description |
|---|---|
src/rfdetr/utilities/reproducibility.py |
Enables PyTorch deterministic algorithms mode with guarded warning behavior; updates reproducibility-related documentation/comments. |
tests/benchmarks/test_training_synthetic.py |
Extends the segmentation convergence benchmark from 5 to 6 epochs and updates the test docstring accordingly. |
Codecov Report❌ Patch coverage is ❌ Your patch check has failed because the patch coverage (60%) is below the target coverage (95%). You can increase the patch coverage or adjust the target coverage. Additional details and impacted files@@ Coverage Diff @@
## develop #1307 +/- ##
=======================================
- Coverage 84% 84% -0%
=======================================
Files 110 110
Lines 13623 13628 +5
=======================================
+ Hits 11454 11457 +3
- Misses 2169 2171 +2 🚀 New features to boost your workflow:
|
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
- Reformatted multi-line comment in `seed_all` to improve clarity and adhere to line length conventions.
Borda
added a commit
that referenced
this pull request
Aug 11, 2026
- seed_all now calls torch.use_deterministic_algorithms(True, warn_only=True) after the cuDNN flags, so every op with a deterministic kernel uses it and the rest warn at execution time instead of raising - guard the call with try/except RuntimeError, logging via get_logger so a failure to enable determinism degrades to a warning rather than propagating out of seed_all - import get_logger in reproducibility - test_train_convergence_segmentation: raise epochs 5 -> 6 and update the docstring epoch references, widening the convergence margin above GPU run-to-run noise --------- Co-authored-by: claude[bot] <209825114+claude[bot]@users.noreply.github.com> Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This pull request improves reproducibility and updates a synthetic training benchmark test. The main changes include making PyTorch's deterministic algorithms setting more robust and updating the segmentation convergence test to run for an additional epoch.
Reproducibility improvements:
seed_allfunction to usetorch.use_deterministic_algorithmswithwarn_only=True, ensuring deterministic operations where possible and logging a warning if full determinism can't be enabled instead of raising an error. [1] [2]Benchmark test updates:
test_train_convergence_segmentationto check results after 6 epochs instead of 5, and to run for 6 epochs, improving test reliability. [1] [2]