enable 3d weights for NVFP4Tensor #3109
Open
Conversation
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
|
Stack from ghstack (oldest at bottom): |
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/3109
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit baf7568 with merge base 8955739 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
bot
added
the
CLA Signed
vkuzo
added
the
topic: improvement
vkuzo
changed the title
jcaip
approved these changes
Oct 2, 2025
| assert len(data_hp.shape) == 2, "unsupported" | ||
| M, K = data_hp.shape[0], data_hp.shape[1] | ||
| assert len(data_hp.shape) in (2, 3), "unsupported" | ||
| leading_dims, M, K = data_hp.shape[:-2], data_hp.shape[-2], data_hp.shape[-1] |
There was a problem hiding this comment.
just fyi, i think you can do:
*leading_dims, M, K = data_hp.shape
| new = NVFP4Tensor( | ||
| new_qdata, | ||
| new_scale, | ||
| old._block_size, |
There was a problem hiding this comment.
would block size change with transpose?
There was a problem hiding this comment.
currently block_size is an integer for this tensor, 16 for NVFP4. If we change it to a multidimensional block, we'd have to update this code.
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Outdated suggestions cannot be applied.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:
Enables NVFP4Tensor to be created from a 3d weight. Note that slicing is gated to 2d tensors for now, we can enable that in a future PR if needed.
This is needed for vLLM stitching 2d weights into a 3d weight for MoEs.
Test Plan:
Reviewers:
Subscribers:
Tasks:
Tags: