transformers
7d3ae42e - Remove sdpa support for models using softcapping (#49811)

Commit
4 days ago
Remove sdpa support for models using softcapping (#49811) * Warn when sdpa or a flash kernel drops the attention softcap * unsupported sdpa * others * no need for tracing check, better to raise there * raise instead * remove test --------- Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com>
Author
Parents
Loading