Update app.py

app.py

@@ -104,24 +104,23 @@ def respond(message, api_key, max_tokens, top_p, temperature):
         response = requests.post(API_URL, headers=headers, data=json.dumps(data), stream=True)
 
         if response.status_code == 200:
-            assistant_reply = ""
             for line in response.iter_lines(decode_unicode=True):
                 if line.strip():
-                    if line.strip() == "[DONE]":
-                     break  
-        
-                    try:
+                     try:
                          if line.startswith("data:"):
-                         line = line[5:].strip()
-
+                           line = line[5:].strip()
+                
                          if line:
-                            chunk = json.loads(line)
-                            chunk_message = chunk.get("chunk", {}).get("content", "")
-                            assistant_reply += chunk_message
-                            yield assistant_reply
-                    except json.JSONDecodeError as e:
-                        print(f"Stream chunk error: {e} with line: {line}")
-
+                             chunk = json.loads(line)
+                             chunk_message = chunk.get("chunk", {}).get("content", "")
+                             assistant_reply += chunk_message
+                
+                             partial_history = history + [(message, assistant_reply, "You", "P-ALPLE", session["avatar"], ASSISTANT_PIC_PATH)]
+                             sessions[api_key]["history"] = partial_history
+                
+                             yield assistant_reply
+                     except json.JSONDecodeError as e:
+                          print(f"Stream chunk error: {e} with line: {line}")
 
 
             history.append((message, assistant_reply, "You", "P-ALPLE", sessions[api_key]["avatar"], ASSISTANT_PIC_PATH))