CalvinBackup/facefusion-labs
1
0
Fork 0
mirror of https://github.com/facefusion/facefusion-labs.git synced 2026-04-19 15:56:37 +02:00
Code Issues Packages Projects Releases Wiki Activity

stabilize training

Browse Source
This commit is contained in:
harisreedhar
2025-06-17 13:06:47 +05:30
parent fc766b8327
commit e894e4172a
3 changed files with 19 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
hyperswap/README.md
+5 -1
View File
@@ -89,9 +89,13 @@ mask_weight = 5.0
```
[training.trainer]
accumulate_size = 4
learning_rate = 0.0004
generator_learning_rate = 0.0004
discriminator_learning_rate = 0.0002
momentum = 0.5
gradient_clip = 20.0
noise_factor = 0.05
scheduler_factor = 0.7
scheduler_patience = 2000
max_epochs = 50
strategy = auto
precision = 16-mixed
hyperswap/config.ini
+5 -1
View File
@@ -47,9 +47,13 @@ mask_weight =
[training.trainer]
accumulate_size =
learning_rate =
generator_learning_rate =
discriminator_learning_rate =
momentum =
gradient_clip =
noise_factor =
scheduler_factor =
scheduler_patience =
max_epochs =
strategy =
precision =
hyperswap/src/training.py
+9 -5
View File
@@ -35,7 +35,11 @@ class HyperSwapTrainer(LightningModule):
self.config_face_masker_path = config_parser.get('training.model', 'face_masker_path')
self.config_noise_factor = config_parser.getfloat('training.trainer', 'noise_factor')
self.config_accumulate_size = config_parser.getfloat('training.trainer', 'accumulate_size')
self.config_learning_rate = config_parser.getfloat('training.trainer', 'learning_rate')
self.config_generator_learning_rate = config_parser.getfloat('training.trainer', 'generator_learning_rate')
self.config_discriminator_learning_rate = config_parser.getfloat('training.trainer', 'discriminator_learning_rate')
self.config_momentum = config_parser.getfloat('training.trainer', 'momentum')
self.config_scheduler_factor = config_parser.getfloat('training.trainer', 'scheduler_factor')
self.config_scheduler_patience = config_parser.getint('training.trainer', 'scheduler_patience')
self.config_gradient_clip = config_parser.getfloat('training.trainer', 'gradient_clip')
self.config_preview_frequency = config_parser.getint('training.trainer', 'preview_frequency')
self.generator_embedder = torch.jit.load(self.config_generator_embedder_path, map_location = 'cpu').eval()
@@ -62,10 +66,10 @@ class HyperSwapTrainer(LightningModule):
return output_tensor, output_mask
def configure_optimizers(self) -> Tuple[OptimizerSet, OptimizerSet]:
generator_optimizer = torch.optim.AdamW(self.generator.parameters(), lr = self.config_learning_rate, betas = (0.5, 0.999), weight_decay = 1e-4, eps = 1e-8)
discriminator_optimizer = torch.optim.AdamW(self.discriminator.parameters(), lr = self.config_learning_rate * 0.5, betas = (0.5, 0.999), weight_decay = 1e-4, eps = 1e-8)
generator_scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(generator_optimizer, mode = 'min', factor = 0.7, patience = 2000, min_lr = 1e-8)
discriminator_scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(discriminator_optimizer, mode = 'min', factor = 0.7, patience = 2000, min_lr = 1e-8)
generator_optimizer = torch.optim.AdamW(self.generator.parameters(), lr = self.config_generator_learning_rate, betas = (self.config_momentum, 0.999), weight_decay = 1e-4, eps = 1e-8)
discriminator_optimizer = torch.optim.AdamW(self.discriminator.parameters(), lr = self.config_discriminator_learning_rate, betas = (self.config_momentum, 0.999), weight_decay = 1e-4, eps = 1e-8)
generator_scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(generator_optimizer, mode = 'min', factor = self.config_scheduler_factor, patience = self.config_scheduler_patience, min_lr = 1e-8)
discriminator_scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(discriminator_optimizer, mode = 'min', factor = self.config_scheduler_factor, patience = self.config_scheduler_patience, min_lr = 1e-8)
generator_config =\
{