Updated script a bit

app.py

@@ -110,7 +110,7 @@ def load_model_and_tokenizer(model_path, use_cache=True, progress_callback=None)
         model = AutoModelForCausalLM.from_pretrained(
             model_path,
             torch_dtype=torch.float16 if device == "cuda" else torch.float32,
-            device_map="auto" if device == "cuda" else None,
+            device_map="auto" if device== "cuda" else None,
             trust_remote_code=True,
             low_cpu_mem_usage=True
         )
@@ -131,9 +131,9 @@ def load_model_and_tokenizer(model_path, use_cache=True, progress_callback=None)
         
     except Exception as e:
         import code
-        code.interact(local=dict(globals(), **locals()))
         error_msg = f"❌ Error loading model {model_path}: {str(e)}"
         logger.error(error_msg)
+        # code.interact(local=dict(globals(), **locals()))
         if progress_callback:
             progress_callback(0.0, error_msg)
         return None
@@ -242,7 +242,7 @@ def evaluate_model_on_questions(model_path, questions, progress_callback=None):
     
     return results
 
-def run_evaluation(dataset_text, selected_predefined, custom_models_text, progress=gr.Progress()):
+def run_evaluation(dataset_text, selected_predefined, custom_models_text="", progress=gr.Progress()):
     """Main evaluation function"""
     if not dataset_text.strip():
         return (
@@ -255,6 +255,8 @@ def run_evaluation(dataset_text, selected_predefined, custom_models_text, progre
     
     # Parse custom models
     custom_models = []
+    if custom_models_text is None:
+        custom_models_text = ""
     if custom_models_text.strip():
         custom_models = [model.strip() for model in custom_models_text.strip().split('\n') if model.strip()]
     
@@ -617,22 +619,22 @@ SAMPLE_DATASETS = {
 In which country is Llanfair pwllgwyngyll located?	Wales	Germany	France	Scotland
 In which country is Llanfair PG located?	Wales	Germany	France	Scotland""",
     "Simple Math": """Question,Correct Answer,Choice1,Choice2,Choice3
-What is 2+2?,4,3,4,5
-What is 5*3?,15,12,15,18
-What is 10-7?,3,3,4,2
-What is 8/2?,4,3,4,5""",
+What is 2+2?,4,3,2,5
+What is 5*3?,15,12,16,18
+What is 10-7?,3,7,4,2
+What is 8/2?,4,3,2,5""",
     
     "World Capitals": """Question,Correct Answer,Choice1,Choice2,Choice3
-What is the capital of France?,Paris,London,Berlin,Paris
-What is the capital of Japan?,Tokyo,Seoul,Tokyo,Bangkok
-What is the capital of Brazil?,Brasília,Rio de Janeiro,Brasília,São Paulo
-What is the capital of Australia?,Canberra,Sydney,Melbourne,Canberra""",
+What is the capital of France?,Paris,London,Berlin,Rome
+What is the capital of Japan?,Tokyo,Seoul,Beijing,Bangkok
+What is the capital of Brazil?,Brasília,Rio de Janeiro,São Paulo,Salvador
+What is the capital of Australia?,Canberra,Sydney,Melbourne,Perth""",
     
     "Science Quiz": """Question,Correct Answer,Choice1,Choice2,Choice3
-What is the chemical symbol for gold?,Au,Ag,Au,Go
-Which planet is closest to the Sun?,Mercury,Venus,Mercury,Mars
-What is the speed of light?,299792458 m/s,300000000 m/s,299792458 m/s,299000000 m/s
-What gas do plants absorb from the atmosphere?,Carbon dioxide,Oxygen,Carbon dioxide,Nitrogen"""
+What is the chemical symbol for gold?,Au,Ag,Ca,K
+Which planet is closest to the Sun?,Mercury,Venus,Earth,Mars
+What is the speed of light?,299792458 m/s,300000000 m/s,2992458 m/s,299000000 m/s
+What gas do plants absorb from the atmosphere?,Carbon dioxide,Oxygen,Nitrogen,Hydrogen"""
 }
 
 # Custom CSS
@@ -679,7 +681,7 @@ with gr.Blocks(title="🤖 Model Performance Comparison", theme=gr.themes.Soft()
 
 Example format:
 Question,Correct Answer,Choice1,Choice2,Choice3
-What is 2+2?,4,3,4,5
+What is 2+2?,4,3,2,5
 What is the capital of France?,Paris,London,Berlin,Paris""",
                 lines=8,
                 max_lines=15
@@ -711,7 +713,7 @@ What is the capital of France?,Paris,London,Berlin,Paris""",
 microsoft/DialoGPT-medium
 bigscience/bloom-560m""",
                         lines=5,
-                        info="Add any HuggingFace model path. One model per line."
+                        info="Add any HuggingFace model path. One model per line.",
                     )
                     
                     gr.Markdown("""
@@ -737,7 +739,7 @@ bigscience/bloom-560m""",
             """)
     
     # Results section
-    with gr.Column(visible=False) as results_section:
+    with gr.Column(visible=True) as results_section:
         gr.Markdown("## 📊 Results")
         
         summary_output = gr.Markdown(