Update app.py

app.py

@@ -121,47 +121,46 @@ with gr.Blocks() as demo:
             if assistant_msg is not None:
                 chat_data.append({"role": "assistant", "content": assistant_msg})
 
-        if refusal_condition(history[-1][0]):
-            history[-1][1] = '抱歉，我無法回答您這個問題'
-            return history
-
         message = tokenizer.apply_chat_template(chat_data, tokenize=False)
         message = message[3:]  # remove SOT token
 
-        
-        data = {
-            "model": MODEL_NAME,
-            "prompt": str(message),
-            "temperature": float(temperature) + 0.01,
-            "n": 1,
-            "max_tokens": int(max_new_tokens),
-            "stop": "",
-            "top_p": float(top_p),
-            "logprobs": 0,
-            "echo": False,
-            "presence_penalty": PRESENCE_PENALTY,
-            "frequency_penalty": FREQUENCY_PENALTY,
-            "stream": True,
-        }
- 
-        with requests.post(API_URL, headers=HEADERS, data=json.dumps(data), stream=True) as r:
-            for response in r.iter_lines():
-                if len(response) > 0:
-                    text = response.decode()
-                    if text != "data: [DONE]":
-                        if text.startswith("data: "):
-                            text = text[5:]
-                        delta = json.loads(text)["choices"][0]["text"]
-
-                        if history[-1][1] is None:
-                            history[-1][1] = delta
-                        else:
-                            history[-1][1] += delta
-                        yield history
-        if history[-1][1].endswith('</s>'):
-            history[-1][1] = history[-1][1][:-4]
+        if refusal_condition(history[-1][0]):
+            history[-1][1] = '抱歉，我無法回答您這個問題'
             yield history
-
+        else:
+            data = {
+                "model": MODEL_NAME,
+                "prompt": str(message),
+                "temperature": float(temperature) + 0.01,
+                "n": 1,
+                "max_tokens": int(max_new_tokens),
+                "stop": "",
+                "top_p": float(top_p),
+                "logprobs": 0,
+                "echo": False,
+                "presence_penalty": PRESENCE_PENALTY,
+                "frequency_penalty": FREQUENCY_PENALTY,
+                "stream": True,
+            }
+     
+            with requests.post(API_URL, headers=HEADERS, data=json.dumps(data), stream=True) as r:
+                for response in r.iter_lines():
+                    if len(response) > 0:
+                        text = response.decode()
+                        if text != "data: [DONE]":
+                            if text.startswith("data: "):
+                                text = text[5:]
+                            delta = json.loads(text)["choices"][0]["text"]
+    
+                            if history[-1][1] is None:
+                                history[-1][1] = delta
+                            else:
+                                history[-1][1] += delta
+                            yield history
+            if history[-1][1].endswith('</s>'):
+                history[-1][1] = history[-1][1][:-4]
+                yield history
+    
         print('== Record ==\nQuery: {query}\nResponse: {response}'.format(query=repr(message), response=repr(history[-1][1])))
 
     msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(