Resolve double BOS token issue #462
Open
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
This disables tokenizer's call to prepend BOS token when tokenizing context which already has BOS token created by its own chat template.
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Outdated suggestions cannot be applied.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.
When using chat template, the BOS token is automatically added to the context. This PR disables adding of the second BOS token, which at the moment happens automatically when the context passes through tokenizer's call which by default is called with
add_special_tokens=True(ref: https://github.com/huggingface/lighteval/blob/ed084813e0bd12d82a06d9f913291fdbee774905/src/lighteval/models/vllm/vllm_model.py#L90).Before this PR, inputs to the model would look like this:
After this PR, inputs to the model would look like this:
Important Note: Right now, there is no support in
lightevalto parse this argument from string input into boolean representation, so this PR will have effect ones huggingface/lighteval#598 is merged.