mirror of
https://github.com/kohya-ss/sd-scripts.git
synced 2026-04-09 06:45:09 +00:00
format by black
This commit is contained in:
@@ -5,21 +5,33 @@ from typing import List, Optional, Union
|
||||
|
||||
|
||||
def apply_snr_weight(loss, timesteps, noise_scheduler, gamma):
|
||||
alphas_cumprod = noise_scheduler.alphas_cumprod
|
||||
sqrt_alphas_cumprod = torch.sqrt(alphas_cumprod)
|
||||
sqrt_one_minus_alphas_cumprod = torch.sqrt(1.0 - alphas_cumprod)
|
||||
alpha = sqrt_alphas_cumprod
|
||||
sigma = sqrt_one_minus_alphas_cumprod
|
||||
all_snr = (alpha / sigma) ** 2
|
||||
snr = torch.stack([all_snr[t] for t in timesteps])
|
||||
gamma_over_snr = torch.div(torch.ones_like(snr)*gamma,snr)
|
||||
snr_weight = torch.minimum(gamma_over_snr,torch.ones_like(gamma_over_snr)).float() #from paper
|
||||
loss = loss * snr_weight
|
||||
return loss
|
||||
alphas_cumprod = noise_scheduler.alphas_cumprod
|
||||
sqrt_alphas_cumprod = torch.sqrt(alphas_cumprod)
|
||||
sqrt_one_minus_alphas_cumprod = torch.sqrt(1.0 - alphas_cumprod)
|
||||
alpha = sqrt_alphas_cumprod
|
||||
sigma = sqrt_one_minus_alphas_cumprod
|
||||
all_snr = (alpha / sigma) ** 2
|
||||
snr = torch.stack([all_snr[t] for t in timesteps])
|
||||
gamma_over_snr = torch.div(torch.ones_like(snr) * gamma, snr)
|
||||
snr_weight = torch.minimum(gamma_over_snr, torch.ones_like(gamma_over_snr)).float() # from paper
|
||||
loss = loss * snr_weight
|
||||
return loss
|
||||
|
||||
|
||||
def add_custom_train_arguments(parser: argparse.ArgumentParser):
|
||||
parser.add_argument("--min_snr_gamma", type=float, default=None, help="gamma for reducing the weight of high loss timesteps. Lower numbers have stronger effect. 5 is recommended by paper. / 低いタイムステップでの高いlossに対して重みを減らすためのgamma値、低いほど効果が強く、論文では5が推奨")
|
||||
parser.add_argument("--weighted_captions", action="store_true", default=False, help="Enable weighted captions in the standard style (token:1.3). No commas inside parens, or shuffle/dropout may break the decoder.")
|
||||
parser.add_argument(
|
||||
"--min_snr_gamma",
|
||||
type=float,
|
||||
default=None,
|
||||
help="gamma for reducing the weight of high loss timesteps. Lower numbers have stronger effect. 5 is recommended by paper. / 低いタイムステップでの高いlossに対して重みを減らすためのgamma値、低いほど効果が強く、論文では5が推奨",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--weighted_captions",
|
||||
action="store_true",
|
||||
default=False,
|
||||
help="Enable weighted captions in the standard style (token:1.3). No commas inside parens, or shuffle/dropout may break the decoder.",
|
||||
)
|
||||
|
||||
|
||||
re_attention = re.compile(
|
||||
r"""
|
||||
|
||||
Reference in New Issue
Block a user