[Bert2Bert] allow bert2bert + relative embeddings #14324
Merged
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
4 tasks
sgugger
approved these changes
Nov 8, 2021
Comment on lines
455
to
459
| # cross attention cannot have relative position embeddings | ||
| cross_attention_config = copy.deepcopy(config) | ||
| cross_attention_config.position_embedding_type = "absolute" | ||
|
|
||
| self.crossattention = BertAttention(cross_attention_config) |
There was a problem hiding this comment.
This is a bit hackish. If we start to turn this argument on and off, maybe it should be an init argument? It can default to None in which case we take the config value (so this is 100% backward compatible).
|
Thanks @patrickvonplaten for fixing this! It looks correct that relative position embeddings should not be used on cross-attention layers. |
patrickvonplaten
commented
Nov 9, 2021
…/patrickvonplaten/transformers into fix_bert2bert_relative_pos_embds
…into fix_bert2bert_relative_pos_embds
|
Merging now to prevent merge conflicts |
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Outdated suggestions cannot be applied.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.
What does this PR do?
Fixes #14010
Everything is explained in #14010. IMO, Bert2Bert like models should not (and also cannot really) make use of positional bias in the cross_attention_layers. The PR forces cross attention layers to always use
"absolute"position encodings.Before submitting
Pull Request section?
to it if that's the case.
documentation guidelines, and
here are tips on formatting docstrings.
Who can review?
Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.