import os
import pickle
import gradio as gr
from openai import OpenAI


client = OpenAI(
    base_url="https://yo4x63mj3sbmgpwc.us-east-1.aws.endpoints.huggingface.cloud/v1/",
    api_key=os.environ.get("hf_token"),
)
with open("./question_undetector.pkl", "rb") as f:
    (vectorizer, model) = pickle.load(f)


def guard_question(question):
    pred = model.predict(vectorizer.transform([question]))
    if pred[0] == 1:
        return True
    else:
        return False


def respond(
    지자체,
    제목,
    질문,
    max_tokens,
    temperature,
    top_p,
):
    if guard_question(질문):
        messages = [{"role": "municipality", "content": 지자체}]
        messages.append({"role": "title", "content": 제목})
        messages.append({"role": "question", "content": 질문})
        response = ""

        chat_completion = client.chat.completions.create(
            model="tgi",
            messages=messages,
            stream=True,
            max_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
        )

        for message in chat_completion:
            token = message.choices[0].delta.content
            if token:
                response += token
                yield response
    else:
        yield "제가 답할 수 있는 질문이 아닌 것 같습니다. 저는 민원 게시글을 처리할 수 있어요."


demo = gr.Interface(
    respond,
    inputs=["textbox", "textbox", "textbox"],
    outputs=["textbox"],
    additional_inputs=[
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=1.0, value=0.4, step=0.05, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.90,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)


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