Using the PyTorch API, how to save and load models?

[zhuyuhang4]

Using the PyTorch API, how to save and load models?

[stes]

Hi @zhuyuhang4 ,

multiple options.

If you want to implement saving/loading logic your self, torch.save and torch.load or pretty much any other serialization approach (e.g. pickle, joblib, ...) will do. You should save model, optimizer, relevant arguments, etc.

To make this a bit more convenient, we implemented loading and saving also directly in the Solver class.

The functions essentially take the state_dict of the solver you are running and serialize these.

If this answers your Q, please mark as an answer :) Otherwise, happy to help further.