Fix indexing of lasttoken pooling for longest sequence #2111
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
fixing last token indexing for sequences that span the entire length
|
Nice work! Also find this bug. |
izhx
reviewed
May 25, 2023
change to seq_len
|
Hello! Thanks to you both for spotting this! You've provided a very elegant fix as well - I appreciate it! I'll merge when this goes green.
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Sequences that span the entire seq length of the batch will have attention mask of all 1's.
The current logic results in retrieving the first token embedding instead of the last since
torch.min(torch.tensor([1,1,1 ... , 1]))will be 0.The new logic uses the
attention_mask.shape[1]-1as the index to retrieve wheremin(attention_mask, 1)==1