Skip to content

Conversation

ArthurZucker
Copy link
Collaborator

@ArthurZucker ArthurZucker commented Jan 3, 2025

The goal of the PR is to allow anyone to use a BPE serialized model into a BacktrackingBpe class. This way you only have to use the correct class to leverage the new method.

Serialization should be of BacktrackingBpe type if we want to save time on initializing the double aho corasicks. TBH I am not sure it's worth it, so we'll see.

Full credits to https://github.com/github/rust-gems

@HuggingFaceDocBuilderDev

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

@ArthurZucker
Copy link
Collaborator Author

TODOs:

  • use a trait for BPETrainer to run on BacktrackingBpe
  • implement a converter directly, instead of the deserialization I am currently doing
  • implement model.use_backtracking_bpe() for BPE that just returns a converted BacktrackingBpe

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants