Skip to content

Fix Flux2 DreamBooth prior preservation prompt repeats #13415

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

Open
azolotenkov wants to merge 1 commit into
base: main
Choose a base branch
from
+16 −8
Open

Fix Flux2 DreamBooth prior preservation prompt repeats #13415

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
12 changes: 8 additions & 4 deletions examples/dreambooth/train_dreambooth_lora_flux2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1740,9 +1740,12 @@ def get_sigmas(timesteps, n_dim=4, dtype=torch.float32):
prompt_embeds = prompt_embeds_cache[step]
text_ids = text_ids_cache[step]
else:
num_repeat_elements = len(prompts)
prompt_embeds = prompt_embeds.repeat(num_repeat_elements, 1, 1)
text_ids = text_ids.repeat(num_repeat_elements, 1, 1)
# With prior preservation, prompt_embeds/text_ids already contain [instance, class] entries,
# while collate_fn orders batches as [inst1..instB, class1..classB]. Repeat each entry along
# dim 0 to preserve that grouping instead of interleaving [inst, class, inst, class, ...].
num_repeat_elements = len(prompts) // 2 if args.with_prior_preservation else len(prompts)
prompt_embeds = prompt_embeds.repeat_interleave(num_repeat_elements, dim=0)
text_ids = text_ids.repeat_interleave(num_repeat_elements, dim=0)

# Convert images to latent space
if args.cache_latents:
Expand Down Expand Up @@ -1809,10 +1812,11 @@ def get_sigmas(timesteps, n_dim=4, dtype=torch.float32):
# Chunk the noise and model_pred into two parts and compute the loss on each part separately.
model_pred, model_pred_prior = torch.chunk(model_pred, 2, dim=0)
target, target_prior = torch.chunk(target, 2, dim=0)
weighting, weighting_prior = torch.chunk(weighting, 2, dim=0)

# Compute prior loss
prior_loss = torch.mean(
(weighting.float() * (model_pred_prior.float() - target_prior.float()) ** 2).reshape(
(weighting_prior.float() * (model_pred_prior.float() - target_prior.float()) ** 2).reshape(
target_prior.shape[0], -1
),
1,
Expand Down
12 changes: 8 additions & 4 deletions examples/dreambooth/train_dreambooth_lora_flux2_klein.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1680,9 +1680,12 @@ def get_sigmas(timesteps, n_dim=4, dtype=torch.float32):
prompt_embeds = prompt_embeds_cache[step]
text_ids = text_ids_cache[step]
else:
num_repeat_elements = len(prompts)
prompt_embeds = prompt_embeds.repeat(num_repeat_elements, 1, 1)
text_ids = text_ids.repeat(num_repeat_elements, 1, 1)
# With prior preservation, prompt_embeds/text_ids already contain [instance, class] entries,
# while collate_fn orders batches as [inst1..instB, class1..classB]. Repeat each entry along
# dim 0 to preserve that grouping instead of interleaving [inst, class, inst, class, ...].
num_repeat_elements = len(prompts) // 2 if args.with_prior_preservation else len(prompts)
prompt_embeds = prompt_embeds.repeat_interleave(num_repeat_elements, dim=0)
text_ids = text_ids.repeat_interleave(num_repeat_elements, dim=0)

# Convert images to latent space
if args.cache_latents:
Expand Down Expand Up @@ -1752,10 +1755,11 @@ def get_sigmas(timesteps, n_dim=4, dtype=torch.float32):
# Chunk the noise and model_pred into two parts and compute the loss on each part separately.
model_pred, model_pred_prior = torch.chunk(model_pred, 2, dim=0)
target, target_prior = torch.chunk(target, 2, dim=0)
weighting, weighting_prior = torch.chunk(weighting, 2, dim=0)

# Compute prior loss
prior_loss = torch.mean(
(weighting.float() * (model_pred_prior.float() - target_prior.float()) ** 2).reshape(
(weighting_prior.float() * (model_pred_prior.float() - target_prior.float()) ** 2).reshape(
target_prior.shape[0], -1
),
1,
Expand Down