GPT2Model model output inconsistency between different transformers versions

Intermediate
6

Really appreciate your help @John6666. It worked after I switched back to the “eager” attention with attn_implementation=“eager”.

1 Like