[HF] Add input embedding argument to HF model #442
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
2015aroras
commented
Feb 9, 2024
| @@ -1137,6 +1137,7 @@ def get_alibi_attention_bias(self, seq_len: int, device: torch.device) -> torch. | |||
| def forward( | |||
| self, | |||
| input_ids: torch.LongTensor, | |||
There was a problem hiding this comment.
Ideally I would make input_ids optional, but this would not be a fully backwards compatible change.
epwalsh
approved these changes
Feb 9, 2024
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Outdated suggestions cannot be applied.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 trainers in HF seem to expect models to accept input embeddings directly as an alternative to input ids. This expectation caused this HF issue.
This PR adds an input embeddings argument to OLMo, and bypasses the input embedding layer when input embeddings are provided.