Skip to content

[KERNELS] Enhance matmul_torch to support K-ragged activations with appropriate assertions and output scaling#10364

Closed
Jokeren wants to merge 1 commit into
mainfrom
keren/fix-kernels
Closed

[KERNELS] Enhance matmul_torch to support K-ragged activations with appropriate assertions and output scaling#10364
Jokeren wants to merge 1 commit into
mainfrom
keren/fix-kernels

Conversation

@Jokeren
Copy link
Copy Markdown
Contributor

@Jokeren Jokeren commented May 23, 2026

No description provided.

@Jokeren Jokeren marked this pull request as ready for review May 24, 2026 01:34
@Jokeren Jokeren requested a review from ptillet as a code owner May 24, 2026 01:34
@Jokeren
Copy link
Copy Markdown
Contributor Author

Jokeren commented May 24, 2026

Hi @roman-openai can you take a look? I'm not very sure why it's broken after recent updates

Comment on lines +801 to +805
if (
a_ragged_metadata is not None
and b_ragged_metadata is None
and a.shape[-1] != b.shape[-2]
):
Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why do we need a new special case?

@Jokeren
Copy link
Copy Markdown
Contributor Author

Jokeren commented May 26, 2026

Close the PR as this is not the right fix

@Jokeren Jokeren closed this May 26, 2026
@Jokeren Jokeren deleted the keren/fix-kernels branch May 26, 2026 14:35
@roman-openai
Copy link
Copy Markdown
Collaborator

#10382 - I think this should fix it, lmk if it breaks again!

@Jokeren
Copy link
Copy Markdown
Contributor Author

Jokeren commented May 27, 2026

Thanks!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants