Update app.py

app.py

@@ -143,6 +143,27 @@ def check_messages(history, message, audio):
         
     return history, gr.MultimodalTextbox(value=None, interactive=False), None
 
+def get_streaming_response(response: requests.Response):
+    for chunk in response.iter_lines():
+        if chunk:
+            data = chunk.decode("utf-8")
+            if data.startswith('data: '):
+                json_str = data[6:]
+                
+                if json_str == '[DONE]':
+                    break
+                    
+                try:
+                    chunk = json.loads(json_str)
+                    delta = chunk.get('choices', [{}])[0].get('delta', {})
+                    new_text = delta.get('content', '')
+                    
+                    if new_text:
+                        yield new_text
+                except (json.JSONDecodeError, IndexError):
+                    print(f"Skipping malformed SSE line: {json_str}")
+                    continue
+
 def bot(
     history: list,
     top_p: float,
@@ -161,6 +182,24 @@ def bot(
     
     msgs = history2messages(history)
 
+    headers = {
+        "Content-Type": "application/json"
+    }
+    
+    payload = {
+        "model": "megrez-moe-waic",
+        "messages": [{'role': 'user', 'content': '你谁'}],
+        "max_tokens": 50,
+        "temperature": max(0.7, 0),
+        "top_p": 0.9,
+        "stream": True
+    }
+    API_URL = "http://8.152.0.142:1002/v1/chat/completions"
+    response = requests.post(API_URL, headers=headers, data=json.dumps(payload), timeout=60, stream=True)
+    for chunk in get_streaming_response(response):
+        print(chunk)
+        # yield chunk
+        time.sleep(0.01)
     API_URL = "http://8.152.0.142:8000/v1/chat"
 
     payload = {