Fix model saving at the end of training

huggingface / distil-whisper

Distilled variant of Whisper for speech recognition. 6x faster, 50% smaller, within 1% word error rate.

MIT License

3.32k stars 238 forks source link

Fix model saving at the end of training #117

Closed eustlb closed 2 months ago

eustlb commented 2 months ago

Fix saving of the model at the end of training: save it after the last evaluation so that there is no need to prepare the model with Accelerate, which was previously causing errors.