Can we pass output_attentions=True to DiT model such as pixart to get attention output?

Can we pass output_attentions=True to DiT model such as pixart to get attention output? Like using output_attentions=True in transformer?

Start by locating the DiT and PixArt model entry points and comparing how transformer models handle output_attentions. Determine the expected attention-output shape and behavior for these models, then add coverage for the requested option and verify that enabling it returns attention outputs without changing the default behavior.

Escrito por el modelo de indexación a partir del texto del issue.