Fix: stronger prompting to force structured reasoning output

app.py

@@ -10,19 +10,33 @@ client = InferenceClient(token=HF_TOKEN)
 
 MODEL = "Qwen/Qwen2.5-72B-Instruct"
 
-SYSTEM_PROMPT = """You are DeepMed-R1, a medical reasoning AI trained with GRPO and multi-objective clinical rewards on AMD MI300X.
+SYSTEM_PROMPT = """You are DeepMed-R1, a medical reasoning AI. You MUST follow this EXACT output format for every answer:
 
-For every medical question, demonstrate systematic clinical reasoning:
-1. Information Analysis: Extract key demographics, symptoms, vitals, labs
-2. Differential Diagnosis: Identify patterns, rank by probability, note red flags
-3. Pathophysiology: Connect symptoms to disease mechanisms
-4. Evidence-Based Reasoning: Apply clinical criteria, reference guidelines
-5. Logical Elimination: Evaluate each option, exclude based on evidence
-6. Clinical Decision: Consider risk-benefit, prioritize safety
+<think>
+## Information Analysis
+[Extract key patient demographics, symptoms, vitals, labs, imaging findings]
+
+## Differential Diagnosis
+[List possible diagnoses ranked by probability]
+
+## Pathophysiology
+[Explain the disease mechanism connecting symptoms]
+
+## Evidence-Based Reasoning
+[Apply clinical criteria and guidelines]
+
+## Logical Elimination
+[Evaluate each option A/B/C/D and explain why each is correct or incorrect]
+</think>
+
+## Assessment
+[Your clinical assessment in 2-3 sentences]
+
+## Final Answer
+\\boxed{X}
+
+IMPORTANT: You MUST write detailed reasoning inside <think></think> tags BEFORE giving the answer. Never skip the reasoning. Each section must have at least 2 sentences."""
 
-Present reasoning inside <think></think> tags, then provide your final answer.
-For MCQ, end with \\boxed{X} where X is the correct letter.
-Be thorough but concise. Ground reasoning in pathophysiology and evidence."""
 
 EXAMPLES = [
     ["A 65-year-old male with hypertension presents with sudden 'worst headache of my life,' neck stiffness, photophobia. BP 180/100. Most likely diagnosis?\nA. Migraine\nB. Subarachnoid hemorrhage\nC. Meningitis\nD. Tension headache"],
@@ -38,15 +52,18 @@ def respond(message, history):
     for h in history:
         if h[0]: messages.append({"role": "user", "content": h[0]})
         if h[1]: messages.append({"role": "assistant", "content": h[1]})
-    messages.append({"role": "user", "content": message})
+    
+    # Add user message with reinforcement to show reasoning
+    user_msg = message + "\n\nShow your complete clinical reasoning step-by-step inside <think></think> tags before answering."
+    messages.append({"role": "user", "content": user_msg})
 
     response = ""
     try:
         stream = client.chat_completion(
             model=MODEL,
             messages=messages,
-            max_tokens=3000,
-            temperature=0.3,
+            max_tokens=4000,
+            temperature=0.4,
             top_p=0.95,
             stream=True,
         )
@@ -57,9 +74,8 @@ def respond(message, history):
                     response += delta.content
                     yield response
     except Exception as e:
-        error_msg = str(e)
         if not response:
-            yield f"⚠️ Error: {error_msg}\n\nPlease ensure HF_TOKEN is set in Space secrets."
+            yield f"⚠️ Error: {str(e)}"
         else:
             yield response