github/ai-toolkit
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Added some experimental loss targets

Browse Source
This commit is contained in:
Jaret Burkett
2026-05-24 14:13:23 -06:00
parent 644a6f9246
commit 12304e170f
2 changed files with 30 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
extensions_built_in/sd_trainer/SDTrainer.py
View File

@@ -778,7 +778,8 @@ class SDTrainer(BaseSDTrainProcess):
loss_per_element = (weighing.float() * (denoised_latents.float() - target.float()) ** 2) loss_per_element = (weighing.float() * (denoised_latents.float() - target.float()) ** 2)
loss = loss_per_element loss = loss_per_element
else: else:
if self.train_config.t0_loss_target: local_loss_scale = 1.0
if self.train_config.t0_loss_target or self.train_config.do_fft_loss:
# do the loss on a stepped timestep 0 prediction # do the loss on a stepped timestep 0 prediction
# doto handle doing priors, preservations, masking, etc # doto handle doing priors, preservations, masking, etc
with torch.no_grad(): with torch.no_grad():
@@ -789,10 +790,27 @@ class SDTrainer(BaseSDTrainProcess):
# min 0.001 # min 0.001
tv = torch.clamp(tv, min=0.001) tv = torch.clamp(tv, min=0.001)
# step latent # step latent, use here or with do_fft_loss
t0 = noisy_latents - tv * noise_pred t0 = noisy_latents - tv * noise_pred
target = batch.latents.detach()
pred = t0 if self.train_config.t0_loss_target:
# replace the loss targets and pred
target = batch.latents.detach()
pred = t0
# handle velocity equiv loss if set. This scales t0 loss to match velocity of flowmatchhing loss
if self.train_config.t0_velocity_equiv_weight:
velocity_equiv_weight = (1.0 / torch.clamp(tv, min=0.1) ** 2)
local_loss_scale = velocity_equiv_weight
if self.train_config.do_fft_loss:
with torch.no_grad():
target_mag = torch.fft.rfft2(batch.latents.to(t0.device).float(), norm="ortho").abs()
pred_mag = torch.fft.rfft2(t0.float(), norm="ortho").abs()
fft_loss = F.mse_loss(pred_mag, target_mag, reduction="none")
if self.train_config.do_fft_velocity_equiv_weight:
velocity_equiv_weight = (1.0 / torch.clamp(tv, min=0.1) ** 2)
fft_loss = fft_loss * velocity_equiv_weight
additional_loss += fft_loss.mean()
if self.train_config.loss_type == "pseudo_huber": if self.train_config.loss_type == "pseudo_huber":
diff = pred.float() - target.float() diff = pred.float() - target.float()
c=0.01 c=0.01
@@ -808,6 +826,8 @@ class SDTrainer(BaseSDTrainProcess):
else: else:
loss = torch.nn.functional.mse_loss(pred.float(), target.float(), reduction="none") loss = torch.nn.functional.mse_loss(pred.float(), target.float(), reduction="none")
loss = loss * local_loss_scale
do_weighted_timesteps = False do_weighted_timesteps = False
if self.sd.is_flow_matching: if self.sd.is_flow_matching:
if self.train_config.linear_timesteps or self.train_config.linear_timesteps2: if self.train_config.linear_timesteps or self.train_config.linear_timesteps2:

5
toolkit/config_modules.py
View File

@@ -503,6 +503,11 @@ class TrainConfig:
# do the loss on a timestep to 0 prediction # do the loss on a timestep to 0 prediction
self.t0_loss_target = kwargs.get('t0_loss_target', False) self.t0_loss_target = kwargs.get('t0_loss_target', False)
self.t0_velocity_equiv_weight = kwargs.get('t0_velocity_equiv_weight', False)
# do additional fft loss
self.do_fft_loss = kwargs.get('do_fft_loss', False)
self.do_fft_velocity_equiv_weight = kwargs.get('do_fft_velocity_equiv_weight', False)
# scale the prediction by this. Increase for more detail, decrease for less # scale the prediction by this. Increase for more detail, decrease for less
self.pred_scaler = kwargs.get('pred_scaler', 1.0) self.pred_scaler = kwargs.get('pred_scaler', 1.0)