Tokenizer: fix issue when decoding a single token at a time #1559
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Outdated suggestions cannot be applied.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.
Hi there 👋
When I was working on Phi-3 integration I've spotted an issue with HF Tokenizer that if to decode a single token at time (like in chat script), then it drops all the spaces and the output looks like a single looooooong string (huggingface/transformers#31643).
So, instead of
This is a test string
the code printsThisisateststring
.As was mentioned in the issue by @itazap, the problem happens with
LlamaTokenizer
and thus affects not only Phi-3, but other models too.This PR applies the hack from Phi-3 to other models.