update submit

src/submission/submit.py

@@ -72,6 +72,8 @@ def evaluate_model_accuracy(model_name, num_examples):
         )
         if torch.cuda.is_available():
             model = model.cuda()  # Move model to GPU if available
+        else:
+            model = model.cpu()
 
         # Load your dataset
         dataset = load_dataset("Omartificial-Intelligence-Space/Arabic_Openai_MMMLU")
@@ -125,11 +127,15 @@ Answer:"""
 
                 # Get the top prediction
                 top_prediction = get_top_prediction(text, tokenizer, model)
+                if top_prediction is None:
+                    print(f"Skipping question due to tokenization issues: {data['Question']}")
+                    continue  # Skip this question if no valid options are found
+
                 is_correct = (top_prediction == data['Answer'])
                 correct_predictions += int(is_correct)
                 total_questions += 1
                 overall_correct_predictions += int(is_correct)
-                overall_total_questions +=1
+                overall_total_questions += 1
 
                 detailed_results.append({
                     'Subject': subject,
@@ -163,6 +169,7 @@ Answer:"""
     except Exception as e:
         return f"Error: {str(e)}", pd.DataFrame(), pd.DataFrame()
 
+
 def add_new_eval(
     model: str,
     base_model: str,