Some issue about training proecss

[HankYang]

The training code only uses the training set, and decides whether to save the model based on the loss of the training set. This is very strange. There is also the risk of being overfitting, right?

if average_loss < best_loss:
    best_loss = average_loss
    save_path = os.path.join(args.work_dir, "models", args.run_name, f"epoch{epoch+1}_sam.pth")
    state = {'model': model.float().state_dict(), 'optimizer': optimizer}
    torch.save(state, save_path)