Is it possible to load TF2 SavedModel format into HuggingFace models?

I am trying to fine-tune mbert. Later, I found Google has released version 4 of mbert which isn’t hosted in HuggingFace hub. It can be found in TF Hub in TF2 SaveModel format though. I can’t find anywhere on how to load this model using transformers. Is it even possible at the moment?

1 Like