Auto-pruning Saved Checkpoints about dreambooth-stable-diffusion

[fabbarix]

Due to the prune_ckpt.py declaring the arg parsed in the global scope, when importing it the main arguments are overridden.
Comment out these lines from prune_ckpt.py:

# parser = argparse.ArgumentParser(description='Pruning')
# parser.add_argument('--ckpt', type=str, default=None, help='path to model ckpt')
# args = parser.parse_args()
# ckpt = args.ckpt

Better yet: it may be wise to move the pruner method in a separate file - in ldm - and then use it from there in prune_ckpt.py - so that we avoid these hacks.

[fabbarix]

Nope - that seems to be called only during errors... Need to dig deeper into this one.

[fabbarix]

Ah! Turns out you can just be nice and ask ModelCheckpoint to "save_weights_only": True. Testing it out now to see if it does what the docs say it should do.

[fabbarix]

It does, it makes the checkpoint 4GB, still pruning it will make it closer to 2.5GB. Looking for hooking into the ModelCheckpoint class and see if we can extend it.