[GH-ISSUE #221] Can fp8_paged_mqa_logits support non-contiguous layout? #124

Closed
opened 2026-10-26 21:13:12 +00:00 by bsakaguchi · 1 comment
Owner

Originally created by @MARD1NO on GitHub (Oct 17, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/221

I notice that there exist an assertion

DG_HOST_ASSERT(fused_kv_cache.stride(1) == head_dim_with_sf);

When we sliced paged kvcache by layer_idx, the sliced paged kvcache may not be contiguous.

Originally created by @MARD1NO on GitHub (Oct 17, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/221 I notice that there exist an assertion ``` DG_HOST_ASSERT(fused_kv_cache.stride(1) == head_dim_with_sf); ``` When we sliced paged kvcache by layer_idx, the sliced paged kvcache may not be contiguous.
Author
Owner

@LyricZhao commented on GitHub (Dec 5, 2025):

cc @zheanxu .

<!-- gh-comment-id:3615927203 --> @LyricZhao commented on GitHub (Dec 5, 2025): cc @zheanxu .
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#124
No description provided.