I set vali_loss to sweep_config metric and used wandb.log({"epoch": epoch+1, "train_loss": train_loss, "vali_loss": vali_loss, "test_loss": test_loss}) in my training function to save the loss information, but after training I went to wandb to check the log and found that the result of sweep shows the loss after the training stop, not the loss of the round of earlystopping. How should I deal with this?