tpu

model_training_nnn/rnn_trainer.py

@@ -46,10 +46,9 @@ class BrainToTextDecoder_Trainer:
             gradient_accumulation_steps=args.get('gradient_accumulation_steps', 1),
             log_with=None,  # We'll use our own logging
             project_dir=args.get('output_dir', './output'),
+            even_batches=False,  # Required for batch_size=None DataLoaders
         )
 
-        # Set even_batches to False after initialization - required for batch_size=None DataLoaders
-        # Note: This may not be settable in all Accelerate versions, but we handle it in DataLoader config
 
         # Trainer fields
         self.args = args