Document that train_new_from_iterator uses BPE for WordPiece#2065
Open
adityasingh2400 wants to merge 1 commit into
Open
Conversation
Users following the HF LLM course hit train_new_from_iterator on a WordPiece tokenizer and silently get a BPE-trained model. ArthurZucker confirmed in huggingface#1777 the discrepancy should be documented. Add a callout in the components doc plus a note on WordPieceTrainer so users know to use the trainer with .train() directly when they want WordPiece. Fixes huggingface#1777
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Closes #1777.
Users following the HF LLM course call
train_new_from_iterator()on a WordPiece tokenizer and silently get a BPE-trained vocabulary, because the library uses BPE internally for that path. @ArthurZucker confirmed in #1777 that this should be documented.This PR adds the clarification in two places:
docs/source-doc-builder/components.mdx: a<Tip warning>callout under the Models section pointing readers toWordPieceTrainer+tokenizer.train(...)when they actually want WordPiece.bindings/python/src/trainers.rs: aNote:block in theWordPieceTrainerPython docstring so the same caveat surfaces in API help.No code changes. I skipped regenerating the
.pyistubs since this is docstring-only and my local toolchain may not match CI; happy to add a stub refresh commit if you want.Test plan
git diffshows only docs/docstring edits, no code changes<Tip>syntax matches existing usage inquicktour.mdxNote:block follows the existingArgs:/Example:style intrainers.rs