Skip to content

Conversation

@ArthurZucker
Copy link
Collaborator

What does this PR do?

Previously, all OOV ( and thus timestamp tokens) outputed by the model are decoded to <|endoftext|> by the xxx.en whisper models. This does not happen with the multilingual model only because I added "" to the vocabulary, and the unk_token_id is the same "". But this does not really make sense.
As the default behavior for Whisper is just to outptu "" for any OOV, now the _convert_id_to_token function does not use a unk_token.

This will fix the inconsistency, and will help for the whisper refactoring.

@HuggingFaceDocBuilderDev
Copy link

HuggingFaceDocBuilderDev commented Jan 23, 2023

The documentation is not available anymore as the PR was closed or merged.

Copy link
Collaborator

@sgugger sgugger left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the fix! Make sure to run make style on your branch to fix the quality issue.

@ArthurZucker ArthurZucker merged commit d8415ba into huggingface:main Jan 23, 2023
Copy link
Contributor

@sanchit-gandhi sanchit-gandhi left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

SGTM aligning with the official implementation here 👍 thanks for the fix!

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.

4 participants