Retrain a model multiple times

Hi everyone,

I was wondering if we can train a model on day 1, save it, then do another training on day 20 but on additional data without losing what the model has already learned. Is it something we can do with PyTorch? If yes, can someone provide a very simple example, like from saving the model to reload, keeping in mind that we can’t lose what the model has already learned?

I hope I was clear, Thank you.


It can be easily done in PyTorch, I recommend you to read the documentation.

Thank you, I was looking exactly for this part :slight_smile

Saving & Loading a General Checkpoint for Inference and/or Resuming Training