How to bring fine-tuned model (no vocab.json) to the correct format

[Heite92]

Hi, I'm kinda lost here, because the model I'm getting from fine tuning seems different from what others get and what the h5-to-ggml script expects.

I used this tutorial: https://medium.com/@shridharpawar77/a-comprehensive-guide-for-custom-data-fine-tuning-with-the-whisper-model-60e4cbce736d

However, it seems pretty identical to https://huggingface.co/blog/fine-tune-whisper

If I understand correctly h5 is a tensorflow 1.x model format.

So far I was able to save the model as a pt-file with torch.save, but that model isn't compatible with the pt-to-ggml script.

I found others with the same problem in the depths of the internet, but none had a solution or atleast I didn't find it.

Anyone has an idea on how to get it to ggml?