Repository navigation
Fix FP8 offload alignment for cuBLAS 12.9/13.0 - #355
Merged
Merged
Conversation
pengcuo
approved these changes
Nov 3, 2026
lfengad
approved these changes
Nov 3, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
TL;DR: Align packed offload tensors to 16 bytes to prevent FP8 failures in the recommended CUDA 13.0 environment.
Follow-up to #11, caught by @satishra-ai in #11 (review). Packing scalar scales at 4-byte offsets triggers a cuBLAS 12.9/13.0 alignment bug and CUBLAS_STATUS_NOT_SUPPORTED in both offload modes. Adds padding and a regression test.
Validation: 70 tests passed on GB300 (PyTorch 2.13/CUDA 13.3); the regression fails before the fix and passes afterward. Lint and formatting pass. H100/CUDA 13.0 was not rerun locally.