After fine-tuning a model (with curie) I found that the F1 validation metric was higher at the end of epoch 3 than it is at the end of fine-tuning (i.e. epoch 4). Using the API, can I recover the state of the fine-tuned model the way it was at the end of epoch3, and use that for inference? I would like to do that without having to re-fine-tune the model from the beginning for 3 epochs.
1 Like