[GH-ISSUE #165] wgrad_gemm_fp8_fp8_fp32_nt not supporting arbitrary m? #91

Closed
opened 2026-10-26 21:12:17 +00:00 by bsakaguchi · 1 comment
Owner

Originally created by @TheTinyTeddy on GitHub (Aug 15, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/165

Many thanks for the great work!

I have used the older version where wgrad_gemm still exists, and I was trying to do a 1*128 128 *1 scaled FP8 MMA with m=1 or something not power of 2 (since this normally is the token dimension where m can vary a lot) such as m=133, and I encountered "illegal memory" error or "too many kernels" error depending on the m.

Anyway to fix this?

Kind regards

Originally created by @TheTinyTeddy on GitHub (Aug 15, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/165 Many thanks for the great work! I have used the older version where wgrad_gemm still exists, and I was trying to do a 1*128 128 *1 scaled FP8 MMA with m=1 or something not power of 2 (since this normally is the token dimension where m can vary a lot) such as m=133, and I encountered "illegal memory" error or "too many kernels" error depending on the m. Anyway to fix this? Kind regards
Author
Owner

@TheTinyTeddy commented on GitHub (Aug 18, 2025):

Okay, I found the later commit works!

<!-- gh-comment-id:3194962104 --> @TheTinyTeddy commented on GitHub (Aug 18, 2025): Okay, I found the later commit works!
Sign in to join this conversation.
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepGEMM#91
No description provided.