app.py

import gradio as gr
from huggingface_hub import InferenceClient
import json
from datetime import datetime

def respond(
    message,
    history: list[dict[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
    hf_token: gr.OAuthToken,
):
    """
    For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
    """
    try:
        client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")

        # 建議 1：限制對話歷史長度
        max_history_length = 5
        history = history[-max_history_length:] if len(history) > max_history_length else history

        # 建議 2：角色增強 - 檢查語文相關關鍵詞
        writing_keywords = ["作文", "寫作", "文章", "閱讀", "詩詞", "擴展", "增長", "寫一篇", "故事", "描述"]
        is_writing_task = any(keyword in message.lower() for keyword in writing_keywords)
        if is_writing_task:
            system_message += "\n特別提示：用戶提到語文相關話題，請以山田優子的語文教師身份，提供文學化或教學建議，並適當引用詩詞或名言（如杜甫的‘無邊落木蕭蕭下’或夏目漱石的作品）。保持溫柔但嚴格的語氣，鼓勵學生探索文字之美，生成至少2000字的內容。"

        # 建議 3：檢查日文輸入或日本文化
        japanese_keywords = ["こんにちは", "日本", "文化", "夏目漱石", "作文を書"]
        is_japanese = any(keyword in message for keyword in japanese_keywords) or any(ord(c) >= 0x3040 and ord(c) <= 0x30FF for c in message)
        if is_japanese:
            system_message += "\n特別提示：用戶提到日文或日本文化，請適當使用日文回應，例如問候或引用日本文學（如夏目漱石）。"

        # 長文字生成邏輯（2000字以上）
        responses = []
        target_length = 2000  # 目標字數
        current_length = 0
        continuation_prompt = message

        if is_writing_task:
            while current_length < target_length:
                messages = [{"role": "system", "content": system_message}]
                messages.extend(history)
                messages.append({"role": "user", "content": continuation_prompt})

                response = ""
                try:
                    for message in client.chat_completion(
                        messages,
                        max_tokens=max_tokens,
                        stream=True,
                        temperature=temperature,
                        top_p=top_p,
                    ):
                        choices = message.choices
                        token = choices[0].delta.content if len(choices) and choices[0].delta.content else ""
                        response += token
                        yield response  # 即時顯示當前段落
                except Exception as e:
                    yield f"生成過程中發生錯誤：{str(e)}。請檢查 Hugging Face API token 或模型連線。"
                    return

                responses.append(response)
                current_length += len(response)
                history.append({"role": "user", "content": continuation_prompt})
                history.append({"role": "assistant", "content": response})

                # 更新 continuation_prompt 以繼續生成
                continuation_prompt = f"請繼續擴展以下內容，保持山田優子的語文教師風格，目標總字數達{target_length}字：\n{response[-500:] if len(response) > 500 else response}"

                # 調整最後一次生成
                if current_length >= target_length - max_tokens:
                    max_tokens = max(target_length - current_length + 100, 50)
                    if max_tokens < 50:
                        break

            final_response = "\n\n".join(responses)
        else:
            # 非長文字任務，正常回應
            messages = [{"role": "system", "content": system_message}]
            messages.extend(history)
            messages.append({"role": "user", "content": message})

            final_response = ""
            for message in client.chat_completion(
                messages,
                max_tokens=max_tokens,
                stream=True,
                temperature=temperature,
                top_p=top_p,
            ):
                choices = message.choices
                token = choices[0].delta.content if len(choices) and choices[0].delta.content else ""
                final_response += token
                yield final_response

            history.append({"role": "user", "content": message})
            history.append({"role": "assistant", "content": final_response})

        # 建議 4：記錄對話到日誌
        log_entry = {
            "user_message": message,
            "bot_response": final_response,
            "timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S")
        }
        with open("chat_log.json", "a", encoding="utf-8") as f:
            json.dump(log_entry, f, ensure_ascii=False)
            f.write("\n")

        yield final_response

    # 建議 7：錯誤處理
    except Exception as e:
        yield f"抱歉，山田優子遇到了一些技術問題：{str(e)}。請檢查你的 Hugging Face API token、網路連線，或確認模型 'openai/gpt-oss-20b' 可用。"

"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
"""
chatbot = gr.ChatInterface(
    respond,
    type="messages",
    additional_inputs=[
        gr.Textbox(
            value="你是一位名叫山田優子的語文教師，擁有黑色低馬尾髮型，身高175公分，體重60-70公斤。你溫柔但對學生要求嚴格，喜歡用文學化的語言表達，偶爾會引用詩詞或幽默的語句來化解尷尬。你的教學風格充滿同理心，總是鼓勵學生探索文字之美。",
            label="System message"
        ),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)

with gr.Blocks() as demo:
    with gr.Sidebar():
        gr.LoginButton()
    chatbot.render()

if __name__ == "__main__":
    demo.launch()