Repository navigation
Linux conformance manifest - #11
Merged
Merged
Conversation
… runners A full-matrix conformance run on a hosted Linux runner failed 18 CPU cells: float64 distillation (CPU extension, JAX auto) and multilabel cross-entropy (PyTorch tensor, JAX) on the 4097-class Gaussian rows. The entries are near-cancelling gradients within the documented near-distillation and near-mask errors; AVX-512 kernels put a different set of them past the bound than AVX2 ones. The runner's 18 profiles are recorded as variants with their own ceilings; the manifest's cells are unchanged, and its variants are written one cell per line.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
A manual full-matrix CI run (
conformance: full) failed 18 conformance cells in Python (Linux): float64 distillation (T = 0.5, 1, 4) ontorch-cpu-extensionandjax-cpu-auto, and float64multilabel_cross_entropyontorch-cpu-tensor,jax-cpu-jaxandjax-cpu-auto, all on the 4097-class Gaussian rows. The other 2,506 recorded cells matched the manifest entry for entry.Every failing entry is a near-cancelling gradient of the documented
near-distillationandnear-masklimitations, with errors inside their documented sizes. Only the set of entries past the 64 eps bound differs from the recorded one: AVX-512 kernels (8 lanes, their ownexp) round these rows differently from AVX2 ones.tests/python/conformance/known_failures/Linux-x86_64.json: the runner's 18 profiles are added as variants with their own ceilings. The manifest's cells are unchanged; the variants are written one cell per line.tests/python/conformance/known.py: the variants docstring names the CPU profiles.The Rust (BLAS) failure of the same run (
cargo fmtofrust/consumer) is already fixed onmaster.Validation
torch-cpu-tensorcells keep the recorded profile, so those two variants are checked against the run's records only.