Skip to content

[mxfp8 training] 128b alignment for CUTensormap to support CUDA 13.0+#3837

Merged
danielvegamyhre merged 1 commit into
mainfrom
alignfeb9
Feb 6, 2026
Merged

[mxfp8 training] 128b alignment for CUTensormap to support CUDA 13.0+#3837
danielvegamyhre merged 1 commit into
mainfrom
alignfeb9

Conversation

@danielvegamyhre

@danielvegamyhre danielvegamyhre commented Feb 6, 2026 •
edited
Loading

Copy link
Copy Markdown
Contributor

This was developed with CUDA 12.8, but the alignment requirement changed in 13.0.

Test

(torch) dev@gpu-dev-b0912ea8:/tmp/ao$ export CUDA_VERSION=13.0
(torch) dev@gpu-dev-b0912ea8:/tmp/ao$ export CUDA_HOME=/usr/local/cuda-13.0     
(torch) dev@gpu-dev-b0912ea8:/tmp/ao$ USE_CPP=1 pip install --no-build-isolation -e ./ -v
  • CUDA_VISIBLE_DEVICES=1 pytest test/prototype/moe_training/test_training.py -v -s

@danielvegamyhre danielvegamyhre added mx topic: bug fix Use this tag for PRs that fix bugs labels Feb 6, 2026
@pytorch-bot

pytorch-bot Bot commented Feb 6, 2026 •
edited
Loading

Copy link
Copy Markdown

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/3837

Note: Links to docs will display an error until the docs builds have been completed.

✅ You can merge normally! (1 Unrelated Failure)

As of commit 8fc4119 with merge base b4ac203 (image):

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

This comment was automatically generated by Dr. CI and updates every 15 minutes.

@meta-cla meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Feb 6, 2026
@slayton58

slayton58 commented Feb 6, 2026

Copy link
Copy Markdown

Smallest of nits: Your 12.9 doc link still says 64B alignment, it went up to 128B with CUDA 13

@danielvegamyhre danielvegamyhre changed the title [mxfp8 training] 128b alignment for CUTensormap to support CUDA 12.9+ [mxfp8 training] 128b alignment for CUTensormap to support CUDA 13.0+ Feb 6, 2026
@danielvegamyhre

danielvegamyhre commented Feb 6, 2026

Copy link
Copy Markdown
Contributor Author

Smallest of nits: Your 12.9 doc link still says 64B alignment, it went up to 128B with CUDA 13

ah, thanks!

@danielvegamyhre

danielvegamyhre commented Feb 6, 2026

Copy link
Copy Markdown
Contributor Author

test failure unrelated, landing

@danielvegamyhre danielvegamyhre merged commit cd1cb37 into main Feb 6, 2026
21 of 22 checks passed
@jerryzh168 jerryzh168 added the module: training quantize_ api training flow label Feb 13, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@slayton58 slayton58 slayton58 approved these changes

Assignees

No one assigned

Labels

CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. module: training quantize_ api training flow mx topic: bug fix Use this tag for PRs that fix bugs

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@danielvegamyhre @slayton58 @jerryzh168