Skip to content

feat(deepseek_v4): 1. use rope_rotate_activation instead of rotary_em…#701

Merged
valarLip merged 1 commit into
feat/deepseek-v4-pr1-skeletonfrom
feat/deepseek-v4-pr1-skeleton_jun_0506
May 6, 2026
Merged

feat(deepseek_v4): 1. use rope_rotate_activation instead of rotary_em…#701
valarLip merged 1 commit into
feat/deepseek-v4-pr1-skeletonfrom
feat/deepseek-v4-pr1-skeleton_jun_0506

Conversation

@junhaha666

Copy link
Copy Markdown
Contributor
  1. use rope_rotate_activation instead of rotary_emb + rotate_activation; related aiter pr add rope/rotate_activation/fp4_quant_inplace fused kernel for dsv4 aiter#3035
  2. use topk_softplus fused kernel; related aiter pr add topk_softplus kernel aiter#2995
  3. use mhc_pre in hc_head; related aiter pr Update mhc_pre hip kernel support hc_head aiter#3044
  4. add scale_indexer_weights

…b + rotate_activation; 2. use topk_softplus fused kernel; 3. use mhc_pre in hc_head; 4. add scale_indexer_weights
@valarLip valarLip merged commit 3d38bdf into feat/deepseek-v4-pr1-skeleton May 6, 2026
@valarLip valarLip deleted the feat/deepseek-v4-pr1-skeleton_jun_0506 branch May 6, 2026 10:33
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.

2 participants