Update utils.py

utils.py

@@ -49,7 +49,7 @@ def postprocess(
         return y
 
 def count_token(input_str):
-    encoding = tiktoken.encoding_for_model("gpt-3.5-turbo")
+    encoding = tiktoken.get_encoding("cl100k_base")
     length = len(encoding.encode(input_str))
     return length
 
@@ -144,14 +144,20 @@ def stream_predict(openai_api_key, system_prompt, history, inputs, chatbot, prev
     try:
         response = get_response(openai_api_key, system_prompt, history, temperature, top_p, True)
     except requests.exceptions.ConnectTimeout:
-        status_text = standard_error_msg + error_retrieve_prompt
+        history.pop()
+        status_text = standard_error_msg + "连接超时，无法获取对话。" + error_retrieve_prompt
+        yield get_return_value()
+        return
+    except requests.exceptions.ReadTimeout:
+        history.pop()
+        status_text = standard_error_msg + "读取超时，无法获取对话。" + error_retrieve_prompt
         yield get_return_value()
         return
 
     chatbot.append((parse_text(inputs), ""))
     yield get_return_value()
 
-    for chunk in tqdm(response.iter_lines()):
+    for chunk in response.iter_lines():
         if counter == 0:
             counter += 1
             continue
@@ -160,7 +166,12 @@ def stream_predict(openai_api_key, system_prompt, history, inputs, chatbot, prev
         if chunk:
             chunk = chunk.decode()
             chunklength = len(chunk)
-            chunk = json.loads(chunk[6:])
+            try:
+                chunk = json.loads(chunk[6:])
+            except json.JSONDecodeError:
+                status_text = f"JSON解析错误。请重置对话。收到的内容: {chunk}"
+                yield get_return_value()
+                break
             # decode each line as response data is in bytes
             if chunklength > 6 and "delta" in chunk['choices'][0]:
                 finish_reason = chunk['choices'][0]['finish_reason']
@@ -169,7 +180,12 @@ def stream_predict(openai_api_key, system_prompt, history, inputs, chatbot, prev
                     print("生成完毕")
                     yield get_return_value()
                     break
-                partial_words = partial_words + chunk['choices'][0]["delta"]["content"]
+                try:
+                    partial_words = partial_words + chunk['choices'][0]["delta"]["content"]
+                except KeyError:
+                    status_text = standard_error_msg + "API回复中找不到内容。很可能是Token计数达到上限了。请重置对话。当前Token计数: " + str(sum(previous_token_count)+token_counter+user_token_count)
+                    yield get_return_value()
+                    break
                 if token_counter == 0:
                     history.append(construct_assistant(" " + partial_words))
                 else: