app.py

import gradio as gr
from huggingface_hub import InferenceClient
import os
from threading import Event

hf_token = os.getenv("HF_TOKEN")
stop_event = Event()

models = {
    "deepseek-ai/DeepSeek-Coder-V2-Instruct": "(한국회사)DeepSeek-Coder-V2-Instruct",
    "meta-llama/Meta-Llama-3.1-8B-Instruct": "Meta-Llama-3.1-8B-Instruct",
    "mistralai/Mixtral-8x7B-Instruct-v0.1": "Mixtral-8x7B-Instruct-v0.1",
    "CohereForAI/c4ai-command-r-plus": "Cohere Command-R Plus"
}

def get_client(model):
    return InferenceClient(model=model, token=hf_token)

MAX_HISTORY_LENGTH = 5  # 히스토리에 유지할 최대 대화 수

def truncate_history(history):
    return history[-MAX_HISTORY_LENGTH:] if len(history) > MAX_HISTORY_LENGTH else history

def respond(message, history, system_message, max_tokens, temperature, top_p, selected_model):
    stop_event.clear()
    client = InferenceClient(model=selected_model, token=hf_token)
    
    truncated_history = truncate_history(history)
    
    messages = [{"role": "system", "content": system_message + "\n사용자의 입력에만 직접적으로 답변하세요. 추가 질문을 생성하거나 사용자의 입력을 확장하지 마세요."}]
    messages.extend([{"role": "user" if i % 2 == 0 else "assistant", "content": m} for h in truncated_history for i, m in enumerate(h) if m])
    messages.append({"role": "user", "content": message})
    
    try:
        response = ""
        for chunk in client.text_generation(
            prompt="\n".join([f"{m['role']}: {m['content']}" for m in messages]),
            max_new_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            stream=True
        ):
            if stop_event.is_set():
                break
            if chunk:
                response += chunk
                if response.startswith(message):
                    response = response[len(message):].lstrip()
                yield truncated_history + [(message, response)]
        
    except Exception as e:
        yield truncated_history + [(message, f"오류 발생: {str(e)}")]

def continue_writing(message, history, system_message, max_tokens, temperature, top_p, selected_model):
    if not history:
        yield [("시스템", "대화 내역이 없습니다.")]
        return

    truncated_history = truncate_history(history)
    last_assistant_message = truncated_history[-1][1]
    
    prompt = f"이전 대화를 간단히 요약하고 이어서 작성해주세요. 마지막 응답: {last_assistant_message[:100]}..."
    
    for response in respond(prompt, truncated_history[:-1], system_message, max_tokens, temperature, top_p, selected_model):
        yield response

def stop_generation():
    stop_event.set()
    return "생성이 중단되었습니다."

def regenerate(chat_history, system_message, max_tokens, temperature, top_p, selected_model):
    if not chat_history:
        return "대화 내역이 없습니다."
    last_user_message = chat_history[-1][0]
    return respond(last_user_message, chat_history[:-1], system_message, max_tokens, temperature, top_p, selected_model)

with gr.Blocks() as demo:
    chatbot = gr.Chatbot()
    msg = gr.Textbox(label="메시지 입력", placeholder="메시지를 입력하세요. Enter로 전송, Shift+Enter로 줄바꿈")

    with gr.Row():
        send = gr.Button("전송")
        continue_btn = gr.Button("계속 작성")
        regenerate_btn = gr.Button("🔄 재생성")
        stop = gr.Button("🛑 생성 중단")
        clear = gr.Button("🗑️ 대화 내역 지우기")

    with gr.Accordion("추가 설정", open=True):
        system_message = gr.Textbox(
            value="너는 나의 최고의 비서이다.\n내가 요구하는것들을 최대한 자세하고 정확하게 답변하라.\n반드시 한글로 답변할것.",
            label="시스템 메시지",
            lines=5
        )
        max_tokens = gr.Slider(minimum=1, maximum=2000, value=500, step=100, label="최대 새 토큰 수")
        temperature = gr.Slider(minimum=0.1, maximum=2.0, value=0.7, step=0.05, label="온도")
        top_p = gr.Slider(minimum=0.1, maximum=1.0, value=0.90, step=0.05, label="Top-p (핵 샘플링)")
        model = gr.Radio(list(models.keys()), value=list(models.keys())[0], label="언어 모델 선택", info="사용할 언어 모델을 선택하세요")

    # Event handlers
    msg.submit(respond, [msg, chatbot, system_message, max_tokens, temperature, top_p, model], [chatbot])
    send.click(respond, [msg, chatbot, system_message, max_tokens, temperature, top_p, model], [chatbot])
    continue_btn.click(continue_writing, 
        inputs=[msg, chatbot, system_message, max_tokens, temperature, top_p, model],
        outputs=[chatbot])
    regenerate_btn.click(regenerate, [chatbot, system_message, max_tokens, temperature, top_p, model], [chatbot])
    stop.click(stop_generation, outputs=[msg])
    clear.click(lambda: None, outputs=[chatbot])

if __name__ == "__main__":
    if not hf_token:
        print("경고: HF_TOKEN 환경 변수가 설정되지 않았습니다. 일부 모델에 접근할 수 없을 수 있습니다.")
    demo.launch()