Skip to content

[fix] [model] auto-patch all Attention layers to ensure cu_seq_lens stays on CPU for NPU fused-attention.#199

Open
A1waysBeenHere wants to merge 1 commit intoByteDance-Seed:mainfrom
A1waysBeenHere:main
Open

[fix] [model] auto-patch all Attention layers to ensure cu_seq_lens stays on CPU for NPU fused-attention.#199
A1waysBeenHere wants to merge 1 commit intoByteDance-Seed:mainfrom
A1waysBeenHere:main

Commits