How can we get the attention weights of the input tokens during the inference [Discussion]

[Occupying-Mars]

Discussion

Been trying to experiment with Llava and i wanted to get attention weights of the input tokens and since it uses llamaspadattetnion + the monkeypatch i wanted to know how we can get the attention weights of the input tokens (both for text and image) during the inference (new and learning about pytorch pardon if this is a dumb issue)