Skip to content

Conversation

@wdziurdz
Copy link
Contributor

@wdziurdz wdziurdz commented Nov 26, 2025

related #5128

@wdziurdz wdziurdz changed the base branch from dev/wdziurdz/test-matmul-1 to main November 26, 2025 15:27
@wdziurdz wdziurdz marked this pull request as draft November 26, 2025 15:27
@wdziurdz wdziurdz force-pushed the dev/wdziurdz/test-matmul-15 branch 2 times, most recently from a8c2529 to 62ea159 Compare November 27, 2025 09:12
@wdziurdz wdziurdz requested a review from anmyachev November 27, 2025 11:38
@wdziurdz wdziurdz force-pushed the dev/wdziurdz/test-matmul-15 branch 2 times, most recently from ddce0c9 to eaee091 Compare November 27, 2025 13:38
@anmyachev
Copy link
Contributor

Just FYI: I'm currently trying to merge the commits needed for this change.

aeng-openai and others added 5 commits November 28, 2025 08:50
any mxfp where natively supported requires using the persistent matmul
kernel. in these cases, do not use heuristics to resolve `is_persistent`

Signed-off-by: Witold Dziurdz <[email protected]>
This PR supports
- the `CDNA4MXScaleLayout.unswizzle_data` method used in GPT-OSS model
- padding tensors with 0 when doing scale preshuffling

Signed-off-by: Witold Dziurdz <[email protected]>
…t None and do_gamma is set

Signed-off-by: Witold Dziurdz <[email protected]>
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.

7 participants