sanchit-gandhi/whisper_event_winners · Post training actions

Sanchit,

Thanks a lot for your work and contribution to the field. Really appreciate.

We do fine-tuning Whisper Large model but for some specific reasons we don't want to push model to the Hub (data sensitivity, specific case). We know this is not the best dev/ops experience but we have to live with it.

What are following actions to "pack" the model after successful training? As you can imagine on training directory we have checkpoints saved ~ last few of them etc.

We're asking this because when we use same model to try to convert it to JAX or other experiments you shared we loose some capabilities, like timestamps etc.

Thanks in advance!