Skip to content

Z-Image: no attention mask if attention mask is a no-op #13184

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
dxqb wants to merge 4 commits into
base: main
Choose a base branch
from
+6 −0
Open

Z-Image: no attention mask if attention mask is a no-op #13184

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
9380e58
Release: v0.36.0-release
sayakpaul Dec 8, 2025
6fa06bd
no mask if no mask needed
dxqb Feb 25, 2026
b17c46d
Revert "Release: v0.36.0-release"
dxqb Feb 25, 2026
492b804
fix
dxqb Feb 25, 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
6 changes: 6 additions & 0 deletions src/diffusers/models/transformers/transformer_z_image.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -580,6 +580,8 @@ def forward(
x_attn_mask = torch.zeros((bsz, x_max_item_seqlen), dtype=torch.bool, device=device)
for i, seq_len in enumerate(x_item_seqlens):
x_attn_mask[i, :seq_len] = 1
if torch.all(x_attn_mask):
x_attn_mask = None

if torch.is_grad_enabled() and self.gradient_checkpointing:
for layer in self.noise_refiner:
Expand Down Expand Up @@ -608,6 +610,8 @@ def forward(
cap_attn_mask = torch.zeros((bsz, cap_max_item_seqlen), dtype=torch.bool, device=device)
for i, seq_len in enumerate(cap_item_seqlens):
cap_attn_mask[i, :seq_len] = 1
if torch.all(cap_attn_mask):
cap_attn_mask = None

if torch.is_grad_enabled() and self.gradient_checkpointing:
for layer in self.context_refiner:
Expand All @@ -633,6 +637,8 @@ def forward(
unified_attn_mask = torch.zeros((bsz, unified_max_item_seqlen), dtype=torch.bool, device=device)
for i, seq_len in enumerate(unified_item_seqlens):
unified_attn_mask[i, :seq_len] = 1
if torch.all(unified_attn_mask):
unified_attn_mask = None

if torch.is_grad_enabled() and self.gradient_checkpointing:
for layer in self.layers:
Expand Down