transformers
Support per-layer cache configuration and attention-mask selection
#47901
Merged

Loading