Closed JackKelly closed 2 years ago
Thomas suggests:
Looking at the workspace, another thing that takes time is uploading your checkpoints. You could use
from pytorch_lightning.callbacks import ModelCheckpoint checkpoint_callback = ModelCheckpoint(monitor='val_accuracy', mode='max')
to filter out checkpoints. To not slow your training, wandb will wait until the training has ended to sync those.
Thomas suggests:
Looking at the workspace, another thing that takes time is uploading your checkpoints. You could use
to filter out checkpoints. To not slow your training, wandb will wait until the training has ended to sync those.