keras-io icon indicating copy to clipboard operation
keras-io copied to clipboard

Fix incorrect reference to model in Transformer class

Open MeysamMoghaddam opened this issue 8 months ago • 1 comments

Description: In the train_step and test_step methods of the Transformer class, there's a reference to model.compute_loss() which should be self.compute_loss() instead. Since we're inside the class methods, we should use self to refer to the current instance of the model.

Files Changed:

transformer_asr.py

Changes Made:

Replaced model.compute_loss() with self.compute_loss() in both train_step and test_step methods

Impact: This correction ensures proper access to the parent class's compute_loss method through the correct instance reference. The original code would raise a NameError since model is not defined in this context, while self properly refers to the current Transformer instance.

MeysamMoghaddam avatar Apr 01 '25 04:04 MeysamMoghaddam

Can you updated the generated files? markdown and ipynb?

mattdangerw avatar May 01 '25 00:05 mattdangerw