init

app.py

@@ -25,7 +25,7 @@ def fine_tune_model(model_name, dataset_name, hub_id, api_key, num_epochs, batch
         model = get_peft_model(model, lora_config)
         tokenizer = AutoTokenizer.from_pretrained(model_name)
     
-        max_length = 256
+        max_length = 64
     
         # Tokenize the dataset
         def tokenize_function(examples):
@@ -63,7 +63,7 @@ def fine_tune_model(model_name, dataset_name, hub_id, api_key, num_epochs, batch
             per_device_eval_batch_size=int(batch_size), 
             num_train_epochs=int(num_epochs),
             weight_decay=0.01,
-            gradient_accumulation_steps=int(grad),
+            #gradient_accumulation_steps=int(grad),
             load_best_model_at_end=True,
             metric_for_best_model="accuracy",
             greater_is_better=True,
@@ -73,7 +73,7 @@ def fine_tune_model(model_name, dataset_name, hub_id, api_key, num_epochs, batch
             hub_model_id=hub_id.strip(),
             fp16=True,
             #lr_scheduler_type='cosine',
-            save_steps=500,  # Save checkpoint every 500 steps
+            save_steps=200,  # Save checkpoint every 500 steps
             save_total_limit=3, 
         )
         # Check if a checkpoint exists and load it