Skip to content

Fix DREAM training #8302

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Jun 1, 2024
Merged

Fix DREAM training #8302

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
d7f1e18
Fix DREAM training
May 28, 2024
fc01c40
Merge branch 'main' into fix-dream-training
sayakpaul May 29, 2024
feba5ec
Merge branch 'main' into fix-dream-training
yiyixuxu May 29, 2024
17db4fa
Merge branch 'main' into fix-dream-training
sayakpaul Jun 1, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 4 additions & 4 deletions src/diffusers/training_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -157,19 +157,19 @@ def compute_dream_and_update_latents(
with torch.no_grad():
pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample

noisy_latents, target = (None, None)
_noisy_latents, _target = (None, None)
if noise_scheduler.config.prediction_type == "epsilon":
predicted_noise = pred
delta_noise = (noise - predicted_noise).detach()
delta_noise.mul_(dream_lambda)
noisy_latents = noisy_latents.add(sqrt_one_minus_alphas_cumprod * delta_noise)
target = target.add(delta_noise)
_noisy_latents = noisy_latents.add(sqrt_one_minus_alphas_cumprod * delta_noise)
_target = target.add(delta_noise)
elif noise_scheduler.config.prediction_type == "v_prediction":
raise NotImplementedError("DREAM has not been implemented for v-prediction")
else:
raise ValueError(f"Unknown prediction type {noise_scheduler.config.prediction_type}")

return noisy_latents, target
return _noisy_latents, _target


def unet_lora_state_dict(unet: UNet2DConditionModel) -> Dict[str, torch.Tensor]:
Expand Down
Loading