Skip to content

multi-GPU VAE Fix for Cosmos 3 #13924

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
atharvajoshi10 wants to merge 1 commit into
base: main
Choose a base branch
from
+2 −2
Open

multi-GPU VAE Fix for Cosmos 3 #13924

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
4 changes: 2 additions & 2 deletions src/diffusers/pipelines/cosmos/pipeline_cosmos3_omni.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -450,9 +450,9 @@ def _encode_video(self, x: torch.Tensor) -> torch.Tensor:
matches Wan2pt2VAEInterface; no autocast (WanVAE was trained with is_amp=False)."""
in_dtype = x.dtype
dtype = self.vae.dtype
mean = self._vae_latents_mean.to(device=x.device, dtype=dtype)
inv_std = self._vae_latents_inv_std.to(device=x.device, dtype=dtype)
raw_mu = retrieve_latents(self.vae.encode(x.to(dtype)), sample_mode="argmax")
mean = self._vae_latents_mean.to(device=raw_mu.device, dtype=dtype)
inv_std = self._vae_latents_inv_std.to(device=raw_mu.device, dtype=dtype)
return ((raw_mu - mean.view(1, -1, 1, 1, 1)) * inv_std.view(1, -1, 1, 1, 1)).to(in_dtype)

def decode_sound(self, latent: torch.Tensor) -> torch.Tensor:
Expand Down
Loading