Fix llama model sdpa attention forward function masking bug when output_attentions=True#30652

Merged

ArthurZucker merged 14 commits intohuggingface:mainfrom Aladoro:fix-llama-mask-output-attn

May 15, 2024

+212-163

Commits on May 6, 2024

ignore unnecessary sdpa mask converter when output_attentions=True
Aladoro
committed
Merge branch 'huggingface:main' into fix-llama-mask-output-attn
Aladoro
authored
Merge branch 'huggingface:main' into fix-llama-mask-output-attn
Aladoro
authored

Commits on May 12, 2024

Merge branch 'huggingface:main' into fix-llama-mask-output-attn
Aladoro
authored
add tests checking sdpa and eager outputs match when output_attentions=True
Aladoro
committed