app.py

import os
import requests
import gradio as gr

# 从环境变量中读取你的 HF API Token
HF_API_TOKEN = os.environ.get("HF_API_TOKEN")
if HF_API_TOKEN is None:
    raise RuntimeError(
        "环境变量 HF_API_TOKEN 未设置，请在 Space 的 Settings -> Variables 中添加一个名为 HF_API_TOKEN 的 Secret。"
    )

# 想使用的模型 ID，可以自行替换为其他支持 Inference API 的模型
# 比如 "meta-llama/Llama-3.2-1B-Instruct"、"Qwen/Qwen2.5-1.5B-Instruct" 等
MODEL_ID = "Qwen/Qwen2.5-1.5B-Instruct"

API_URL = f"https://api-inference.huggingface.co/models/{MODEL_ID}"
HEADERS = {"Authorization": f"Bearer {HF_API_TOKEN}"}


def query_hf_api(prompt: str, max_new_tokens: int = 256, temperature: float = 0.7) -> str:
    payload = {
        "inputs": prompt,
        "parameters": {
            "max_new_tokens": max_new_tokens,
            "temperature": temperature,
            "do_sample": True,
        },
    }
    response = requests.post(API_URL, headers=HEADERS, json=payload, timeout=120)
    response.raise_for_status()
    data = response.json()

    # text-generation 类模型常见返回格式是 [{"generated_text": "..."}]
    if isinstance(data, list) and len(data) > 0:
        return data[0].get("generated_text", "").strip()
    # 兜底：直接把返回内容转成字符串方便调试
    return str(data)


def chat_fn(history, message, max_new_tokens, temperature):
    # 简单地把历史对话拼成一个长 prompt
    dialog = ""
    if history:
        for user_msg, bot_msg in history:
            dialog += f"用户: {user_msg}\n助手: {bot_msg}\n"
    dialog += f"用户: {message}\n助手:"

    try:
        output = query_hf_api(dialog, max_new_tokens=int(max_new_tokens), temperature=float(temperature))
    except Exception as e:
        output = f"[调用模型出错] {type(e).__name__}: {e}"

    history = history + [(message, output)]
    return history, ""


with gr.Blocks() as demo:
    gr.Markdown(f"# 云端模型聊天 Demo\n使用模型：`{MODEL_ID}`（通过 Hugging Face Inference API）")

    with gr.Row():
        with gr.Column(scale=3):
            chatbot = gr.Chatbot(label="对话", height=500)
            msg = gr.Textbox(
                label="你的问题",
                placeholder="输入你想问的问题，回车或点击发送",
                lines=2,
            )
            send_btn = gr.Button("发送")
            clear_btn = gr.Button("清空对话")

        with gr.Column(scale=1):
            gr.Markdown("### 参数设置")
            max_new_tokens = gr.Slider(16, 512, value=256, step=16, label="max_new_tokens")
            temperature = gr.Slider(0.0, 1.5, value=0.7, step=0.05, label="temperature")

    send_btn.click(
        chat_fn,
        inputs=[chatbot, msg, max_new_tokens, temperature],
        outputs=[chatbot, msg],
    )
    msg.submit(
        chat_fn,
        inputs=[chatbot, msg, max_new_tokens, temperature],
        outputs=[chatbot, msg],
    )
    clear_btn.click(lambda: ([], ""), None, [chatbot, msg])

if __name__ == "__main__":
    # 不要给 launch() 传额外参数，HF 会自己管理 host/port
    demo.launch()