Update app.py

app.py

@@ -5,11 +5,8 @@ from datasets import load_dataset
 import random
 import re
 
-# Load model and tokenizer
-model_name = "abaryan/BioXP-0.5B-MedMCQA"
-
 SYSTEM_PROMPT = """
-You are a medical expert. Answer the medical question with careful analysis and explain why the selected option is correct in 200 words without repeating.
+You are a medical expert. Answer the medical question with careful analysis and explain why the selected option is correct in 2 sentences without repeating.
 Respond in the following format:
 <answer>
 [correct answer]
@@ -19,10 +16,9 @@ Respond in the following format:
 </reasoning>
 """
 
+model_name = "abaryan/BioXP-0.5B-MedMCQA"
 model = AutoModelForCausalLM.from_pretrained(model_name)
 tokenizer = AutoTokenizer.from_pretrained(model_name)
-
-# Load dataset
 dataset = load_dataset("openlifescienceai/medmcqa")
 
 # Move model to GPU if available
@@ -49,11 +45,9 @@ def predict(question: str, option_a: str = "", option_b: str = "", option_c: str
            temperature: float = 0.6, top_p: float = 0.9, max_tokens: int = 256):
     
     # Determine if this is an MCQ by checking if any option is provided
-    # Only treat as MCQ if at least one option is non-empty
     is_mcq = any(opt.strip() for opt in [option_a, option_b, option_c, option_d])
     
     if is_mcq:
-        # Format MCQ question with only non-empty options
         options = []
         if option_a.strip(): options.append(f"A. {option_a}")
         if option_b.strip(): options.append(f"B. {option_b}")
@@ -67,16 +61,12 @@ def predict(question: str, option_a: str = "", option_b: str = "", option_c: str
         formatted_question = f"Question: {question}"
         system_prompt = SYSTEM_PROMPT
     
-    # Create chat-style prompt
     prompt = [
         {'role': 'system', 'content': system_prompt},
         {'role': 'user', 'content': formatted_question}
     ]
     
-    # Use apply_chat_template for better formatting
-    text = tokenizer.apply_chat_template(prompt, tokenize=False, add_generation_prompt=True)
-    
-    # Tokenize and generate
+    text = tokenizer.apply_chat_template(prompt, tokenize=False, add_generation_prompt=True)    
     model_inputs = tokenizer([text], return_tensors="pt").to(device)
     
     with torch.inference_mode():
@@ -87,7 +77,6 @@ def predict(question: str, option_a: str = "", option_b: str = "", option_c: str
             top_p=top_p,
         )
     
-    # Get only the generated response
     generated_ids = generated_ids[0, model_inputs.input_ids.shape[1]:]
     model_response = tokenizer.decode(generated_ids, skip_special_tokens=True)
     
@@ -99,22 +88,21 @@ def predict(question: str, option_a: str = "", option_b: str = "", option_c: str
     # Format output with evaluation if available (only for MCQs)
     output = cleaned_response
     
-    if is_mcq and correct_option is not None:
-        correct_letter = chr(65 + correct_option)
-        answer_match = re.search(r"Answer:\s*([A-D])", cleaned_response, re.IGNORECASE)
-        model_answer = answer_match.group(1).upper() if answer_match else "Not found"
+    # if is_mcq and correct_option is not None:
+    #     correct_letter = chr(65 + correct_option)
+    #     answer_match = re.search(r"Answer:\s*([A-D])", cleaned_response, re.IGNORECASE)
+    #     model_answer = answer_match.group(1).upper() if answer_match else "Not found"
         
-        is_correct = model_answer == correct_letter
-        output += f"\n\n---\nEvaluation:\n"
-        output += f"Correct Answer: {correct_letter}\n"
-        output += f"Model's Answer: {model_answer}\n"
-        output += f"Result: {'✅ Correct' if is_correct else '❌ Incorrect'}\n"
-        if explanation:
-            output += f"\nExpert Explanation:\n{explanation}"
+    #     is_correct = model_answer == correct_letter
+    #     output += f"\n\n---\nEvaluation:\n"
+    #     output += f"Correct Answer: {correct_letter}\n"
+    #     output += f"Model's Answer: {model_answer}\n"
+    #     output += f"Result: {'✅ Correct' if is_correct else '❌ Incorrect'}\n"
+    #     if explanation:
+    #         output += f"\nExpert Explanation:\n{explanation}"
     
     return output
 
-# Create Gradio interface with mobile-optimized design
 with gr.Blocks(
     title="BioXP Medical MCQ Assistant",
     theme=gr.themes.Soft(
@@ -132,7 +120,6 @@ with gr.Blocks(
     
     with gr.Row():
         with gr.Column(scale=1):
-            # Input fields with mobile-friendly spacing
             question = gr.Textbox(
                 label="Medical Question",
                 placeholder="Enter your medical question here...",
@@ -141,7 +128,6 @@ with gr.Blocks(
                 elem_classes=["mobile-input"]
             )
             
-            # Options in a mobile-friendly accordion
             with gr.Accordion("Options", open=True):
                 option_a = gr.Textbox(
                     label="Option A",
@@ -168,7 +154,6 @@ with gr.Blocks(
                     elem_classes=["mobile-input"]
                 )
             
-            # Generation parameters in a collapsible section
             with gr.Accordion("Advanced Settings", open=False):
                 with gr.Row():
                     with gr.Column(scale=1):
@@ -202,13 +187,11 @@ with gr.Blocks(
             correct_option = gr.Number(visible=False)
             expert_explanation = gr.Textbox(visible=False)
             
-            # Buttons with mobile-friendly spacing
             with gr.Row():
                 predict_btn = gr.Button("Get Answer", variant="primary", size="lg", elem_classes=["mobile-button"])
                 random_btn = gr.Button("Random Question", variant="secondary", size="lg", elem_classes=["mobile-button"])
         
         with gr.Column(scale=1):
-            # Output with mobile-friendly styling
             output = gr.Textbox(
                 label="Model's Response",
                 lines=12,
@@ -232,10 +215,8 @@ with gr.Blocks(
         outputs=[question, option_a, option_b, option_c, option_d, correct_option, expert_explanation]
     )
 
-    # Add mobile-optimized CSS
     gr.HTML("""
     <style>
-        /* Mobile-friendly base styles */
         .container {
             max-width: 100%;
             padding: 0.5rem;
@@ -258,7 +239,6 @@ with gr.Blocks(
             font-weight: 500;
         }
         
-        /* Response box styling */
         .response-box {
             font-family: 'Inter', sans-serif;
             line-height: 1.6;