Hi, I am running hyperparameter sweeps with W&B, and training agents like this:
wandb.agent(
args.sweep_id,
function=functools.partial(run_train, args),
project=args.wandb_project,
count=args.max_num_runs,
)
Where run_train uses a pytorch-lightning trainer to train a model. However, my model runs create huge artifacts by seeming to synch a model to the W&B app everytime evaluation happens. I do not understand why it is happening or how to make it stop?
Can someone show me how to stop synching massive artifacts?