update stream

app.py

@@ -30,10 +30,8 @@ def respond(
 ):
     messages = [{"role": "system", "content": system_message}]
 
-    # 添加历史消息
     messages.extend(history)
     
-    # 添加当前用户消息
     messages.append({"role": "user", "content": message})
 
     headers = {
@@ -43,7 +41,7 @@ def respond(
 
     data = {
         "model": "/data/DMind-1-mini",
-        "stream": False,
+        "stream": True,
         "messages": messages,
         "temperature": temperature,
         "top_p": top_p,
@@ -52,33 +50,37 @@ def respond(
         "max_tokens": 16384
     }
 
-    print(f"[INFO] process user msg...")
-    print(f"[INFO] userMsg: {message}")
+    # print(f"[INFO] process user msg...")
+    # print(f"[INFO] userMsg: {message}")
 
     try:
-        with requests.post(API_URL, headers=headers, json=data) as r:
-            # print(f"[INFO] response status: {r.status_code}")
+        with requests.post(API_URL, headers=headers, json=data, stream=True) as r:
             if r.status_code == 200:
-                json_response = r.json()
-                # print(f"[INFO] response json: {json_response}")
-                if 'choices' in json_response and len(json_response['choices']) > 0:
-                    content = json_response['choices'][0].get('message', {}).get('content', '')
-                    # print(f"[INFO] response content: {content}")
-                    if content:
-                        if '<think>' in content and '</think>' in content:
-                            content = content.split('</think>')[-1].strip()
-                        if '**Final Answer**' in content:
-                            content = content.split('**Final Answer**')[-1].strip()
-                        print(f"[INFO] final response: {content}")
-                        return content
-                else:
-                    print(f"[ERROR] No choices in response: {json_response}")
+                current_response = ""
+                for line in r.iter_lines():
+                    if line:
+                        line = line.decode('utf-8')
+                        if line.startswith('data: '):
+                            try:
+                                json_response = json.loads(line[6:])
+                                if 'choices' in json_response and len(json_response['choices']) > 0:
+                                    delta = json_response['choices'][0].get('delta', {})
+                                    if 'content' in delta:
+                                        content = delta['content']
+                                        if content:
+                                            content = content.replace('<', '&lt;').replace('>', '&gt;')
+                                            content = content.replace('*', '\\*')
+                                            current_response += content
+                                            yield current_response
+                            except json.JSONDecodeError:
+                                continue
+                # print(f"[INFO] final response: {current_response}")
             else:
                 print(f"[ERROR] Bad status code: {r.status_code}, response: {r.text}")
-            return "Service temporarily unavailable"
+                yield "Service temporarily unavailable"
     except Exception as e:
         print(f"[ERROR] Request error: {e}")
-        return "Service error occurred"
+        yield "Service error occurred"
 
 
 """