Refactor trainer configuration in train.py for improved clarity. Clean up comments and ensure consistent formatting in evaluation strategy and model selection parameters.

train.py

@@ -192,14 +192,14 @@ def create_trainer(
             model=model,
             tokenizer=tokenizer,
             train_dataset=dataset["train"],
-            eval_dataset=dataset["validation"],  # Add validation dataset
+            eval_dataset=dataset["validation"],
             dataset_text_field="text",
             max_seq_length=max_seq_length,
             dataset_num_proc=2,
             packing=False,
             args=TrainingArguments(
                 per_device_train_batch_size=2,
-                per_device_eval_batch_size=2,  # Add evaluation batch size
+                per_device_eval_batch_size=2,
                 gradient_accumulation_steps=16,
                 warmup_steps=100,
                 max_steps=120,
@@ -207,9 +207,7 @@ def create_trainer(
                 fp16=not is_bfloat16_supported(),
                 bf16=is_bfloat16_supported(),
                 logging_steps=1,
-                evaluation_strategy="steps",  # Add evaluation strategy
                 eval_steps=10,  # Evaluate every 10 steps
-                save_strategy="steps",
                 save_steps=30,
                 save_total_limit=2,
                 optim="adamw_8bit",
@@ -218,9 +216,9 @@ def create_trainer(
                 seed=3407,
                 output_dir="outputs",
                 gradient_checkpointing=True,
-                load_best_model_at_end=True,  # Load best model at the end
-                metric_for_best_model="eval_loss",  # Use validation loss for model selection
-                greater_is_better=False,  # Lower loss is better
+                load_best_model_at_end=True,
+                metric_for_best_model="eval_loss",
+                greater_is_better=False,
             ),
         )
         logger.info("Trainer created successfully")