Repository navigation
Fix/attention mask mismatch - #4218
Closed
KatharinaJacoby wants to merge 4 commits into
Closed
KatharinaJacoby wants to merge 4 commits into
KatharinaJacoby wants to merge 4 commits into
Conversation
attention_mask is truncated to match the sequence length of per_token_entropy. This should fix the tensor mismatch error.
Collaborator
|
Hey @KatharinaJacoby I think another PR that I opened last week resolves the issue that your PR covers too. #4196 |
Member
|
closing in favor of #4196, thanks a lot @KatharinaJacoby and @pramodith |
This file contains hidden or 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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The issue seems to come from the concatenation of the virtual attention mask, which changes the dimensions of the attention_mask and causes a mismatch with per_token_entropy.
Checking that the dimensions of per_token_entropy and attention_mask match before multiplication. I changed the code block so the attention_mask is truncated to match the sequence length of per_token_entropy.
And added a unit test:
-set up common test data in the setUp method.
-test the scenario where the attention mask and per-token entropy have matching dimensions.
-test the scenario where the attention mask and per-token entropy have mismatched dimensions.
-test the scenario where position_ids is used instead of attention_mask.
-test the scenario where neither attention_mask nor position_ids is present, expecting a ValueError.