Skip to content

Fix CrossAttention._sliced_attention #563

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
ydshieh merged 1 commit into from
Sep 19, 2022
Merged

Fix CrossAttention._sliced_attention #563

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: 3 additions & 1 deletion src/diffusers/models/attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -249,13 +249,15 @@ def reshape_batch_dim_to_heads(self, tensor):
return tensor

def forward(self, hidden_states, context=None, mask=None):
batch_size, sequence_length, dim = hidden_states.shape
batch_size, sequence_length, _ = hidden_states.shape

query = self.to_q(hidden_states)
context = context if context is not None else hidden_states
key = self.to_k(context)
value = self.to_v(context)

dim = query.shape[-1]
Copy link
Contributor Author

@ydshieh ydshieh Sep 19, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The dim should be computed with the projected q, k, v (i.e. inner_dim), not with hidden_states.

Copy link
Contributor Author

@ydshieh ydshieh Sep 19, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The current CI passes, but the test has inner_dim == hidden_states.shape[-1] (both 64). When this is not the case, we get an error.


query = self.reshape_heads_to_batch_dim(query)
key = self.reshape_heads_to_batch_dim(key)
value = self.reshape_heads_to_batch_dim(value)
Expand Down