binary-husky/gpt_academic · error · RuntimeError

程序终止。

Error message

程序终止。

What it means

Watchdog abort for the PanGu-Alpha stream: inside the loop over pangu_glm_handle.stream_chat, if time.time() - observe_window[1] exceeds watch_dog_patience (5s) between yields, RuntimeError('程序终止。') ends generation. Local JittorLLMs inference frequently exceeds 5s between tokens on CPU or long prefills.

Source

Thrown at request_llms/bridge_jittorllms_pangualpha.py:136

        if len(observe_window) >= 1: observe_window[0] = load_message + "\n\n" + pangu_glm_handle.info
        if not pangu_glm_handle.success:
            error = pangu_glm_handle.info
            pangu_glm_handle = None
            raise RuntimeError(error)

    # jittorllms 没有 sys_prompt 接口，因此把prompt加入 history
    history_feedin = []
    for i in range(len(history)//2):
        history_feedin.append([history[2*i], history[2*i+1]] )

    watch_dog_patience = 5 # 看门狗 (watchdog) 的耐心, 设置5秒即可
    response = ""
    for response in pangu_glm_handle.stream_chat(query=inputs, history=history_feedin, system_prompt=sys_prompt, max_length=llm_kwargs['max_length'], top_p=llm_kwargs['top_p'], temperature=llm_kwargs['temperature']):
        print(response)
        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, llm_kwargs, plugin_kwargs, chatbot, history=[], system_prompt='', stream = True, additional_fn=None):
    """
        单线程方法
        函数的说明请见 request_llms/bridge_all.py
    """
    chatbot.append((inputs, ""))

    global pangu_glm_handle
    if pangu_glm_handle is None:
        pangu_glm_handle = GetGLMHandle()
        chatbot[-1] = (inputs, load_message + "\n\n" + pangu_glm_handle.info)
        yield from update_ui(chatbot=chatbot, history=[])
        if not pangu_glm_handle.success:
            pangu_glm_handle = None

View on GitHub (pinned to d6bde0fa54)

Solutions

1. Check the JittorLLMs child process is alive and its stderr for crashes/OOM.
2. Increase watch_dog_patience (>=60s for CPU) or update observe_window[1] per yield.
3. Move to GPU / smaller model to keep inter-token latency low.
4. Restart the app to recreate a hung handle if the subprocess is dead.

import time
window = ['', time.time()]
# refresh window[1] each poll; keep patience >= worst-case inter-token gap

try:
    resp = pangu_predict_no_ui_long_connection(..., observe_window=window)
except RuntimeError as e:
    if '程序终止' in str(e):
        window[1] = time.time()
        resp = pangu_predict_no_ui_long_connection(..., observe_window=window)

When it happens

Trigger: PanGu local generation stalls >5s (CPU inference, long prompt prefill, cold loader, or the child process hung/crashed mid-stream).

Common situations: CPU-only device, GPU busy/OOM, subprocess died silently so the generator never yields again, heavily loaded machine.

Related errors

• 程序终止。
• 程序终止。
• 不能正常加载jittorllms的参数！
• error
• 程序终止。

AI-assisted analysis of binary-husky/gpt_academic@d6bde0fa54 (2026-08-14). Data as JSON: /api/errors/6929917ceb234200. Report an issue: GitHub.