Decoding improvements #1033
Merged
Decoding improvements #1033
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
jongwook
changed the title
zackees
pushed a commit
to zackees/whisper
that referenced
this pull request
May 5, 2023
* suppress task tokens (transcribe/translate) * not ignoring the last segment ending with one timestamp
ilanit1997
pushed a commit
to ilanit1997/whisper
that referenced
this pull request
May 16, 2023
* suppress task tokens (transcribe/translate) * not ignoring the last segment ending with one timestamp
abyesilyurt
pushed a commit
to abyesilyurt/whisper
that referenced
this pull request
Nov 13, 2023
* suppress task tokens (transcribe/translate) * not ignoring the last segment ending with one timestamp
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Outdated suggestions cannot be applied.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.
This fixes the failure cases where the model samples the task tokens (
<|transcribe|>or<|translate|>) with a low probability but as reproducibly observed in #1009.Additionally, the segmentation logic is improved to include the last segment in a
DecodingResulteven if it ends with a single timestamp.In a future PR, we may also consider suppressing language tokens when it's not directly after a
<|startoftext|>token