Skip to content

[None][fix] Fix passing scaled timestep to time_embedder in Cosmos3 #15545

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
chang-l merged 2 commits into from
Jun 24, 2026
Merged

[None][fix] Fix passing scaled timestep to time_embedder in Cosmos3 #15545

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
8ac72ae
Fix timestep passed to time_embedder
bastefaniak Jun 23, 2026
460b989
Merge branch 'main' into bstefaniak/fix_cosmos3_visual_gen_timestep
bastefaniak Jun 23, 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
3 changes: 2 additions & 1 deletion tensorrt_llm/_torch/visual_gen/models/cosmos3/pipeline_cosmos3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -586,7 +586,8 @@ def forward_fn(
"""
noise_pred = self.transformer(
hidden_states=latent_input,
timestep=timestep / self.scheduler.config.num_train_timesteps,
timestep=timestep,
attention_timestep=timestep / self.scheduler.config.num_train_timesteps,
text_ids=extra_tensors["text_ids"],
text_mask=extra_tensors["text_mask"],
video_shape=video_shape,
Expand Down
9 changes: 6 additions & 3 deletions tensorrt_llm/_torch/visual_gen/models/cosmos3/transformer_cosmos3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -866,6 +866,7 @@ def forward(
self,
hidden_states: torch.Tensor,
timestep: Optional[torch.Tensor] = None,
attention_timestep: Optional[torch.Tensor] = None,
text_ids: Optional[torch.Tensor] = None,
text_mask: Optional[torch.Tensor] = None,
video_shape: Optional[Tuple[int, int, int]] = None,
Expand All @@ -878,7 +879,9 @@ def forward(

Args:
hidden_states: [B, C, T, H, W] noisy latents
timestep: Normalized diffusion timestep in [0, 1], shape [B]
timestep: Raw scheduler diffusion timestep, shape [B]
attention_timestep: Normalized diffusion timestep in [0, 1], shape [B],
for attention backends that use timestep-dependent behavior.
text_ids: [B, S_text] tokenized text input
text_mask: [B, S_text] attention mask for text (1=real, 0=pad)
video_shape: (T, H, W) in latent space
Expand Down Expand Up @@ -931,7 +934,7 @@ def forward(
text_ids,
text_mask,
freqs_und,
timestep=timestep,
timestep=attention_timestep,
)
self.cached_freqs_gen = freqs_gen

Expand Down Expand Up @@ -971,7 +974,7 @@ def forward(
k_und,
v_und,
freqs_gen,
timestep=timestep,
timestep=attention_timestep,
)

hidden_gen = self.sharder.gather(hidden_gen, dim=1, unpad_to=S_gen)
Expand Down
Loading