Repository navigation
feat: Allow text-only instruct for Voxtral models - #236
Open
tachyonicbytes wants to merge 1 commit into
Open
tachyonicbytes wants to merge 1 commit into
tachyonicbytes wants to merge 1 commit into
Conversation
This branch has not been deployed
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Allow transcribe.cpp to run only instruct, no audio, for models that support that. Currently just the two Voxtral non-streaming models.
This is in order to enable Handy in a separate PR to support post-processing with the same local model used for transcribing, with not API calls. Will work only for Voxtral for now.
Scope
Voxtral's model
src/arch/voxtral/model.cpp, and the more generalsrc/transcribe-asr.cpp.AI Assistance
I have used a model to make the entire feature (including Handy), tested it end to end to work, and then carved it up into the suitable smaller PRs, for further discussion.
Validation
I built the entire post-processing with the same model feature in Handy, and then tested it to work. I solved the remaining bugs and the result seems to be in good condition, good for further investigations. I do not think some needs to test WER and Numerical Validation.
I tried to keep this as minimal as possible, but this is my first attempted change at
transcribe.cpp, so extra care is needed from the maintainers.