dev

app.py

@@ -54,7 +54,7 @@ def fine_tune_model(model, dataset_name, hub_id, api_key, num_epochs, batch_size
         # Load the model and tokenizer
              
         
-    
+        print(f"BATCH: {batch_size}")
         # Set training arguments
         training_args = TrainingArguments(
             output_dir='/data/results',
@@ -68,15 +68,15 @@ def fine_tune_model(model, dataset_name, hub_id, api_key, num_epochs, batch_size
             gradient_accumulation_steps=int(grad),
             #max_grad_norm = 3.0, 
             load_best_model_at_end=True,
-            metric_for_best_model="accuracy",
-            greater_is_better=True,
+            #metric_for_best_model="accuracy",
+            #greater_is_better=True,
             logging_dir='/data/logs',
             logging_steps=10,
             #push_to_hub=True,
             hub_model_id=hub_id.strip(),
             fp16=True,
             #lr_scheduler_type='cosine',
-            save_steps=100,  # Save checkpoint every 500 steps
+            save_steps=500,  # Save checkpoint every 500 steps
             save_total_limit=3, 
         )
         # Check if a checkpoint exists and load it