binary-husky/gpt_academic · warning · RuntimeError

程序终止。

Error message

程序终止。

What it means

Watchdog in predict_no_ui_long_connection for local models: observe_window[1] holds the last 'feed' timestamp; if the caller (UI) stops refreshing it for more than watch_dog_patience (5 s), the generator assumes the user cancelled and raises RuntimeError('程序终止。'). This is cooperative cancellation, not a crash — the model thread simply stops streaming.

Source

Thrown at request_llms/local_llm_class.py:262

                    what_gpt_answer = {}
                    what_gpt_answer["role"] = "assistant"
                    what_gpt_answer["content"] = history[index+1]
                    if what_i_have_asked["content"] != "":
                        if what_gpt_answer["content"] == "":
                            continue
                        history_feedin.append(what_i_have_asked)
                        history_feedin.append(what_gpt_answer)
                    else:
                        history_feedin[-1]['content'] = what_gpt_answer['content']

        watch_dog_patience = 5  # 看门狗 (watchdog) 的耐心, 设置5秒即可
        response = ""
        for response in _llm_handle.stream_chat(query=inputs, history=history_feedin, max_length=llm_kwargs['max_length'], top_p=llm_kwargs['top_p'], temperature=llm_kwargs['temperature']):
            if len(observe_window) >= 1:
                observe_window[0] = response
            if len(observe_window) >= 2:
                if (time.time()-observe_window[1]) > watch_dog_patience:
                    raise RuntimeError("程序终止。")
        return response

    def predict(inputs:str, llm_kwargs:dict, plugin_kwargs:dict, chatbot:ChatBotWithCookies,
                history:list=[], system_prompt:str='', stream:bool=True, additional_fn:str=None):
        """
            refer to request_llms/bridge_all.py
        """
        chatbot.append((inputs, ""))

        _llm_handle = GetSingletonHandle().get_llm_model_instance(LLMSingletonClass)
        chatbot[-1] = (inputs, load_message + "\n\n" + _llm_handle.get_state())
        yield from update_ui(chatbot=chatbot, history=[])
        if not _llm_handle.running:
            raise RuntimeError(_llm_handle.get_state())

        if additional_fn is not None:
            from core_functional import handle_core_functionality
            inputs, history = handle_core_functionality(

View on GitHub (pinned to d6bde0fa54)

Solutions

1. If cancellation was intended, no fix is needed — this is by design.
2. If generation aborts unexpectedly, ensure the caller/UI keeps setting observe_window[1] = time.time() at least every few seconds while the response is still wanted.
3. Pass an observe_window list of length 1 (observation only, no watchdog) when programmatic long calls don't need cancellation.
4. For very slow models, understand the watchdog measures caller liveness, not model progress — do not 'fix' by ignoring it.

Example fix

# caller: keep the watchdog fed while consuming the stream
# before
for response in predict_no_ui_long_connection(inputs, llm_kwargs, observe_window=obs):
    ...

# after
obs[1] = time.time()
for response in predict_no_ui_long_connection(inputs, llm_kwargs, observe_window=obs):
    obs[1] = time.time()  # feed the watchdog each iteration
    ...

# pass a one-element window to run without the watchdog
observe_window = ['']           # no index 1 -> cancellation check disabled
# or keep feeding it:
observe_window = ['', time.time()]

try:
    for r in predict_no_ui_long_connection(q, kw, observe_window=obs):
        obs[1] = time.time()
except RuntimeError as e:
    if str(e) == '程序终止。':
        handle_user_cancel()  # expected, not an error
    else:
        raise

When it happens

Trigger: Streaming from a local model while the front end stops updating observe_window[1] = time.time() for over 5 seconds: user pressed stop, the UI tab was closed/refreshed, or the calling plugin never maintains the second observe_window slot.

Common situations: User clicks 'Stop' during a slow local-model generation; gradio page closed mid-generation; a custom caller passes observe_window of length >= 2 but never refreshes index 1.

Related errors

• 用户取消了程序。
• 程序终止。
• 用户取消了程序。
• 用户取消了程序。
• 用户取消了程序。

AI-assisted analysis of binary-husky/gpt_academic@d6bde0fa54 (2026-08-14). Data as JSON: /api/errors/88b5c11defed6626. Report an issue: GitHub.