Is there a way to get the hidden representations of all 12 layers?

#5
by kmuraish - opened

Hi,

I noticed that the dimensionality of mode_output.hidden_states is [1, {context_size}, 768], which I expected to be [12, {context_size}, 768] like GPT models. I'm assuming it only returns the representation of the last hidden layer.
I would like to know how to obtain the hidden representations in all 12 layers instead of only the last layer.

I appreciate your help!

Sign up or log in to comment