Skip to content

Fix VAE offload encode device mismatch in DreamBooth scripts #13417

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
sayakpaul merged 2 commits into from
Apr 6, 2026
Merged

Fix VAE offload encode device mismatch in DreamBooth scripts #13417

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
68fa726
Fix VAE offload encode device mismatch in DreamBooth scripts
azolotenkov Mar 30, 2026
192b53a
Merge branch 'main' into fix-flux2-vae-offload-device-mismatch
sayakpaul Apr 6, 2026
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
4 changes: 2 additions & 2 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 @@ -1749,8 +1749,8 @@ def get_sigmas(timesteps, n_dim=4, dtype=torch.float32):
model_input = latents_cache[step].mode()
else:
with offload_models(vae, device=accelerator.device, offload=args.offload):
pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
model_input = vae.encode(pixel_values).latent_dist.mode()
pixel_values = batch["pixel_values"].to(device=accelerator.device, dtype=vae.dtype)
model_input = vae.encode(pixel_values).latent_dist.mode()

model_input = Flux2Pipeline._patchify_latents(model_input)
model_input = (model_input - latents_bn_mean) / latents_bn_std
Expand Down
9 changes: 4 additions & 5 deletions examples/dreambooth/train_dreambooth_lora_flux2_img2img.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1686,11 +1686,10 @@ def get_sigmas(timesteps, n_dim=4, dtype=torch.float32):
cond_model_input = cond_latents_cache[step].mode()
else:
with offload_models(vae, device=accelerator.device, offload=args.offload):
pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
cond_pixel_values = batch["cond_pixel_values"].to(dtype=vae.dtype)

model_input = vae.encode(pixel_values).latent_dist.mode()
cond_model_input = vae.encode(cond_pixel_values).latent_dist.mode()
pixel_values = batch["pixel_values"].to(device=accelerator.device, dtype=vae.dtype)
cond_pixel_values = batch["cond_pixel_values"].to(device=accelerator.device, dtype=vae.dtype)
model_input = vae.encode(pixel_values).latent_dist.mode()
cond_model_input = vae.encode(cond_pixel_values).latent_dist.mode()

# model_input = Flux2Pipeline._encode_vae_image(pixel_values)

Expand Down
4 changes: 2 additions & 2 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 @@ -1689,8 +1689,8 @@ def get_sigmas(timesteps, n_dim=4, dtype=torch.float32):
model_input = latents_cache[step].mode()
else:
with offload_models(vae, device=accelerator.device, offload=args.offload):
pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
model_input = vae.encode(pixel_values).latent_dist.mode()
pixel_values = batch["pixel_values"].to(device=accelerator.device, dtype=vae.dtype)
model_input = vae.encode(pixel_values).latent_dist.mode()

model_input = Flux2KleinPipeline._patchify_latents(model_input)
model_input = (model_input - latents_bn_mean) / latents_bn_std
Expand Down
9 changes: 4 additions & 5 deletions examples/dreambooth/train_dreambooth_lora_flux2_klein_img2img.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1634,11 +1634,10 @@ def get_sigmas(timesteps, n_dim=4, dtype=torch.float32):
cond_model_input = cond_latents_cache[step].mode()
else:
with offload_models(vae, device=accelerator.device, offload=args.offload):
pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
cond_pixel_values = batch["cond_pixel_values"].to(dtype=vae.dtype)

model_input = vae.encode(pixel_values).latent_dist.mode()
cond_model_input = vae.encode(cond_pixel_values).latent_dist.mode()
pixel_values = batch["pixel_values"].to(device=accelerator.device, dtype=vae.dtype)
cond_pixel_values = batch["cond_pixel_values"].to(device=accelerator.device, dtype=vae.dtype)
model_input = vae.encode(pixel_values).latent_dist.mode()
cond_model_input = vae.encode(cond_pixel_values).latent_dist.mode()

model_input = Flux2KleinPipeline._patchify_latents(model_input)
model_input = (model_input - latents_bn_mean) / latents_bn_std
Expand Down
4 changes: 2 additions & 2 deletions examples/dreambooth/train_dreambooth_lora_z_image.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1665,8 +1665,8 @@ def get_sigmas(timesteps, n_dim=4, dtype=torch.float32):
model_input = latents_cache[step].mode()
else:
with offload_models(vae, device=accelerator.device, offload=args.offload):
pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
model_input = vae.encode(pixel_values).latent_dist.mode()
pixel_values = batch["pixel_values"].to(device=accelerator.device, dtype=vae.dtype)
model_input = vae.encode(pixel_values).latent_dist.mode()

model_input = (model_input - vae_config_shift_factor) * vae_config_scaling_factor
# Sample noise that we'll add to the latents
Expand Down
Loading