stable-diffusion-finetune/scripts/prune-ckpt.py

48 lines
1.5 KiB
Python
Raw Normal View History

2022-06-11 12:35:03 +02:00
import os
import torch
import fire
2022-08-01 00:13:50 +02:00
def prune_it(p, keep_only_ema=False):
2022-06-11 12:35:03 +02:00
print(f"prunin' in path: {p}")
size_initial = os.path.getsize(p)
nsd = dict()
sd = torch.load(p, map_location="cpu")
print(sd.keys())
for k in sd.keys():
if k != "optimizer_states":
nsd[k] = sd[k]
else:
print(f"removing optimizer states for path {p}")
2022-07-22 11:50:39 +02:00
if "global_step" in sd:
print(f"This is global step {sd['global_step']}.")
2022-08-01 00:13:50 +02:00
if keep_only_ema:
sd = nsd["state_dict"].copy()
# infer ema keys
ema_keys = {k: "model_ema." + k[6:].replace(".", "") for k in sd.keys() if k.startswith("model.")}
new_sd = dict()
for k in sd:
if k in ema_keys:
new_sd[k] = sd[ema_keys[k]]
elif not k.startswith("model_ema.") or k in ["model_ema.num_updates", "model_ema.decay"]:
new_sd[k] = sd[k]
assert len(new_sd) == len(sd) - len(ema_keys)
nsd["state_dict"] = new_sd
fn = f"{os.path.splitext(p)[0]}-pruned.ckpt" if not keep_only_ema else f"{os.path.splitext(p)[0]}-ema-pruned.ckpt"
2022-06-11 12:35:03 +02:00
print(f"saving pruned checkpoint at: {fn}")
torch.save(nsd, fn)
newsize = os.path.getsize(fn)
2022-08-01 00:13:50 +02:00
MSG = f"New ckpt size: {newsize*1e-9:.2f} GB. " + \
f"Saved {(size_initial - newsize)*1e-9:.2f} GB by removing optimizer states"
if keep_only_ema:
MSG += " and non-EMA weights"
print(MSG)
2022-06-11 12:35:03 +02:00
if __name__ == "__main__":
fire.Fire(prune_it)
2022-07-22 11:50:39 +02:00
print("done.")