Enabling the retrieval of whisper's hidden states#1751
Conversation
Hguimaraes
force-pushed
the
whisper_hidden_states
branch
from
2c99032 to
2dcdea6
Compare
|
Hi @Adel-Moumen, Please let me know if there is anything else I could do on this PR. I think I have answered the previous questions and the interface is similar to the wav2vec2 class. Best regards, |
Hello @Hguimaraes, I will finish the review this week :-) |
Adel-Moumen
left a comment
Adel-Moumen
left a comment
There was a problem hiding this comment.
Hello,
Thanks again for your great work! Please take a look at my latest comments :-)
Btw, I wish you a great end of the year :-)
|
|
||
| if self.output_all_hiddens: | ||
| logits, attn = self.forward_decoder( | ||
| out_encoder[-1, ...], decoder_input_ids |
There was a problem hiding this comment.
From my understanding, out_encoder[-1, ...] is exactly the same as out_encoder[-1], so why adding "..." ?
|
Hi @Adel-Moumen, Implemented the changes. If there is anything else, just let me know, and I can update it. Best, |
Adel-Moumen
left a comment
Adel-Moumen
left a comment
There was a problem hiding this comment.
Everything sounds good! Many thanks for the PR! :-)
2 participants
Hello guys,
I'm trying to extend the Whisper class to enable the retrieval of the Hidden States. I think it is handy for people working with speech representation learning. It is a small change. Also, I tried to make it compatible with the previous version, so it should not break any examples or codes that are already using this class.
Best regards,
Heitor