import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import gradio as gr

model = AutoModelForCausalLM.from_pretrained(
    "rinna/bilingual-gpt-neox-4b-instruction-ppo",
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, use_fast=False)

device = model.device


def generate(user_question,
             temperature=0.3,
    system_prompt_template = "システム: もちろんやで！どんどん質問してな！今日も気分ええわ！"

    # one-shot



    user_sample = "ユーザー:日本一の高さの山は？ "
    system_sample = "システム: 富士山や！最高の眺めを拝めるで！！"


    user_prerix = "ユーザー: "
    system_prefix = "システム: "

    output = tokenizer.decode(tokens[0], skip_special_tokens=True)
    return output[len(prompt):]





with gr.Blocks() as demo:
    chat_history = gr.Chatbot()
    inputs = gr.Textbox(label="Question:", placeholder="質問を入力してください")
    outputs = gr.Textbox(label="Answer:")
    btn = gr.Button("Send")
    clear = gr.ClearButton([inputs, chat_history])

    # ボタンが押された時の動作を以下のように定義する:

    btn.click(fn=generate, inputs=inputs, outputs=outputs)

    def response(user_message, chat_history):
        chat_history.append((user_message, system_message))
        return "", chat_history

    inputs.submit(response, inputs=[inputs, chat_history], outputs=[inputs, chat_history])

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