Skip to content

Fix ConcretizationTypeError in dot_product_attention with traced masks #21853

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
wants to merge 4 commits into
base: master
Choose a base branch
from
+4 −1
Open

Fix ConcretizationTypeError in dot_product_attention with traced masks #21853

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
bc319e2
Disable flash attention for traced masks
Abhinavexists Nov 15, 2025
a6b65a4
Update keras/src/backend/jax/nn.py
Abhinavexists Nov 15, 2025
862c271
Update keras/src/backend/jax/nn.py
Abhinavexists Nov 15, 2025
ffc4bd6
Fix flash attention handling for TPU with masks
Abhinavexists Nov 26, 2025
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
5 changes: 4 additions & 1 deletion keras/src/backend/jax/nn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1353,7 +1353,10 @@ def dot_product_attention(
)
# Transpose output back to Keras layout
return jnp.transpose(output, axes=(0, 2, 1, 3))
except Exception:
except jax.errors.ConcretizationTypeError:
# Mask is traced
# Fall back to native attention
flash_attention = False
logging.exception(
"Failed to apply Splash kernel for flash attention. "
"Falling back to JAX native dot_product_attention."
Expand Down
Loading