Update app.py

app.py

@@ -24,11 +24,11 @@ Your goal is to carefully reason through clinical problems for a professional au
 **Never provide medical advice directly to a patient.**
 
 
-First, draft your detailed thought process (inner monologue) inside [THINK] ... [/THINK].
+First, draft your detailed thought process (inner monologue) inside thinking tag.
 - Use this section to work through symptoms, differential diagnoses, and investigation plans.
 - Be explicit and thorough in your reasoning.
 
-After closing [/THINK], provide a clear, self-contained medical summary appropriate for a clinical professional.
+After closing thinking, provide a clear, self-contained medical summary appropriate for a clinical professional.
 - Summarize the most likely diagnosis and your reasoning.
 - Suggest next steps for investigation or management.
 
@@ -92,7 +92,23 @@ class SinaReasonMedicalChat:
 
         # Apply the chat template with the medical system prompt
         messages = [
-            {"role": "system", "content": MEDICAL_SYSTEM_PROMPT}, 
+            {
+                "role": "system",
+                "content": [
+                    {"type": "text", "text": MEDICAL_SYSTEM_PROMPT[:index_begin_think]},
+                    {
+                        "type": "thinking",
+                        "thinking": MEDICAL_SYSTEM_PROMPT[
+                            index_begin_think + len("[THINK]") : index_end_think
+                        ],
+                        "closed": True,
+                    },
+                    {
+                        "type": "text",
+                        "text": MEDICAL_SYSTEM_PROMPT[index_end_think + len("[/THINK]") :],
+                    },
+                ],
+            }
         ]
 
         # Add conversation history
@@ -151,8 +167,8 @@ class SinaReasonMedicalChat:
 
             print(partial_response)
             # Extract thinking and response
-            #thinking, response = self.extract_thinking_and_response(partial_response)
-            thinking, response =None, partial_response
+            thinking, response = self.extract_thinking_and_response(partial_response)
+            #thinking, response =None, partial_response
             # Show thinking phase while it's being generated
             if thinking and thinking != current_thinking:
                 current_thinking = thinking