Function stream_single

inference/app_stream.py:224–249

(prompt, sys_prompt, temperature, max_new_tokens, device,
                  model_path, device_map, torch_dtype, state)

    return state

def stream_single(prompt, sys_prompt, temperature, max_new_tokens, device,
                  model_path, device_map, torch_dtype, state):
    if not prompt or not str(prompt).strip():
        yield "", "请先输入提示词。", state
        return

    t0 = time.time()
    state = ensure_enhancer(state, model_path, device_map, torch_dtype)
    enhancer = state["enhancer"]

    emitted = ""
    try:
        for chunk in enhancer.predict_stream(
            prompt_cot=prompt,
            sys_prompt=sys_prompt,
            temperature=temperature,
            max_new_tokens=max_new_tokens,
            device=device
        ):
            emitted = chunk
            info = f"已接收 {len(emitted)} 字符，用时 {time.time()-t0:.2f}s"
            yield emitted, info, state
        # 结束时再给一次最终状态（可选）
        yield emitted, f"完成。总耗时 {time.time()-t0:.2f}s", state
    except Exception as e:
        yield "", f"推理失败：{e}", state


# 示例数据