binary-husky/gpt_academic · error · RuntimeError

程序终止。

Error message

程序终止。

What it means

Watchdog abort for the MOSS stream: inside the loop over moss_handle.stream_chat, if time.time() - observe_window[1] exceeds watch_dog_patience (5s) without a yield, RuntimeError('程序终止。') is raised. Local MOSS inference between tokens can easily exceed 5s, especially on CPU or with long contexts.

Source

Thrown at request_llms/bridge_moss.py:200

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

    # chatglm 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 moss_handle.stream_chat(query=inputs, history=history_feedin, sys_prompt=sys_prompt, 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, 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 moss_handle
    if moss_handle is None:
        moss_handle = GetGLMHandle()
        chatbot[-1] = (inputs, load_message + "\n\n" + moss_handle.info)
        yield from update_ui(chatbot=chatbot, history=[])
        if not moss_handle.success:
            moss_handle = None

View on GitHub (pinned to d6bde0fa54)

Solutions

1. Check the MOSS thread/process is alive and stderr for OOM/crash.
2. Increase watch_dog_patience (>=60s) or refresh observe_window[1] per yield in the caller.
3. Use GPU/quantization to cut inter-token latency.
4. Restart to rebuild a dead handle before retrying.

import time
window = ['', time.time()]
# refresh window[1] per poll; set patience to hardware worst case

try:
    resp = moss_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 = moss_predict_no_ui_long_connection(..., observe_window=window)

When it happens

Trigger: MOSS local generation stalls >5s between yields: slow CPU inference, long prefill, warm-up after load, or the loader thread hung/crashed mid-stream (e.g. CUDA OOM).

Common situations: CPU-only device, GPU OOM killing generation, first token after cold load, loaded machine, patience left at default 5s.

Related errors

• 程序终止。
• 程序终止。
• 程序终止。
• 程序终止。
• 程序终止。

AI-assisted analysis of binary-husky/gpt_academic@d6bde0fa54 (2026-08-14). Data as JSON: /api/errors/5eefa2d9b218fdcd. Report an issue: GitHub.