Closed kylebgorman closed 4 months ago
I also remove some seemingly pointless logging of model paths; can restore if there's a reason we need this.
Note that there exist alternatives: for instance we can log any checkpoint that's an improvement. From the docs:
# log model only if `val_accuracy` increases wandb_logger = WandbLogger(log_model="all") checkpoint_callback = ModelCheckpoint(monitor="val_accuracy", mode="max") trainer = Trainer(logger=wandb_logger, callbacks=[checkpoint_callback])
But I think this would still be excessive and I can't imagine why you'd want to log a checkpoint on W&B, tbh.
Thanks for catching this. I am sure this was here from copy/pasting from W&B examples. LGTM, pending the one inline comment.
I also remove some seemingly pointless logging of model paths; can restore if there's a reason we need this.
Note that there exist alternatives: for instance we can log any checkpoint that's an improvement. From the docs:
But I think this would still be excessive and I can't imagine why you'd want to log a checkpoint on W&B, tbh.