Smollm2 as language model

[bil-ash]

I would like to train a smaller ultravox model with whisper-small as audio encoder and smollm2 as language model. In the file https://github.com/fixie-ai/ultravox/blob/main/ultravox/training/configs/tiny_config.yaml if I change the text_model property to smollm2-360m-instruct , will that be sufficient or do I need to make other changes also?

[bil-ash]

@farzadab @zqhuang211

[zqhuang211]

Typically that's sufficient. We know some folks have tried smollm2 with ultravox. If it turns out some small changes are needed, please create a PR.