How do you convert pre-trained models to HF standard?

#4
by morenolq - opened

Hi,

I saw that pre trained AST models are released publicly and follow-up models follow the same structure of the backbone model (e.g., https://github.com/YuanGongND/ssast). Is there any way to convert them to match the HF model format? Is this what the HF team did?

Thanks

It's probably here
https://github.com/huggingface/transformers/blob/main/src/transformers/models/audio_spectrogram_transformer/convert_audio_spectrogram_transformer_original_to_pytorch.py

If the follow-up methods use the same model, it will be great if the HF team can do the conversion to publish the weights under the MIT org!

Sign up or log in to comment