[GH-ISSUE #243] Question about k_grouped_gemm #138

Open
opened 2026-10-26 21:13:33 +00:00 by bsakaguchi · 0 comments
Owner

Originally created by @wukong1992 on GitHub (Dec 25, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/243

Why, when computing wgrad in k_grouped, do dy and dx use per_channel_cast_to_fp8 instead of per_token_cast_to_fp8?

Originally created by @wukong1992 on GitHub (Dec 25, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/243 Why, when computing wgrad in k_grouped, do dy and dx use per_channel_cast_to_fp8 instead of per_token_cast_to_fp8?
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#138
No description provided.