How to resume from checkpoint on the hub? Not using trainer api, I'm using TF api

I’m not using the trainer api. I’m using tensorflow api mainly TFAutoModelForSequenceClassification. Is there a way for me to resume from the last pushed checkpoint? I know you can do it with the trainer api.

Hey @cruiser :slight_smile:

Sadly, I think you have to do it manually – i.e. save your model (including the optimizer) using native TF functions, then push all resulting files to the hub.

We have some functionality for Functional/Sequential Keras models (here), but transformer models are Subclassed models.