transformers
[Qwen3.5] Fix GDN linear attention multi-token cached forward
#45513
Merged
Merged
[Qwen3.5] Fix GDN linear attention multi-token cached forward #45513
vasqu
merged
9 commits into
huggingface:main
from
kashif:fix-qwen35-linear-attn-multi-token-cached
Fix Qwen3.5 linear attention multi-token cached forward
49a614ca
Merge branch 'main' into fix-qwen35-linear-attn-multi-token-cached
8c1317b6
vasqu
commented
on 2026-04-23
Review feedback: unify cached-forward state flag, gate single-token/c…
e395f5ac
Propagate linear-attention multi-token cached-forward fix to qwen3_ne…
29a76b36
kashif
changed the title
Merge branch 'main' into fix-qwen35-linear-attn-multi-token-cached
983aedf6
Merge branch 'main' into fix-qwen35-linear-attn-multi-token-cached
23c2b84a
Merge branch 'main' into fix-qwen35-linear-attn-multi-token-cached
8d4864bd
Merge branch 'main' into fix-qwen35-linear-attn-multi-token-cached
cec6c2fb
Review feedback: keep "prefill mode / multi-token decode" comment, la…
40f471b9
Login to write a write a comment.
Login via GitHub