Repository navigation
Compare the guards by steps taken, not wall time - #307
Merged
Merged
Conversation
The sonic-guard test failed on the 2026-10-28 scheduled run (3.12): 5.76 s guarded against 5.20 s unguarded. Each config compiles its own program, and the solves themselves take about 0.3 s either way, so the timing was a comparison of two JIT compiles. Steps are deterministic and show the guard's effect directly: 307 against the 30,000 budget for the sonic guard, 144 against 200,000 for the collapse floor, which had the same check.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The sonic-guard test failed on today's scheduled run (3.12 job, run 40683776219):
assert 5.755954749000011 < 5.203594478999989. Nothing changed on master since the passing run on 10-27.The test timed each
simulatecall end to end. Each config compiles its own program (the guard keys the compile cache), and the solves themselves take about 0.3 s either way (SolverStats.elapsed_s), so the assertion compared two JIT compiles on a shared runner.This compares
SolverStats.nfev(diffraxnum_steps) from the raisedSimulationErrorinstead. Locally:The collapse-floor test had the same wall-clock check. Its 200k-step grind kept the margin wide enough that it had not failed, but it gets the same change.
pytest -m 'slow or not slow' tests/test_validation_refusals.py: 60 passed.ruff check .andtools/pyright_baseline.pyare clean.