app.py

import gradio as gr
from transformers import LlamaTokenizer, LlamaForCausalLM
import torch

model_repo_id = "Bllossom/llama-3-Korean-Bllossom-70B"

# 토크나이저 로드
tokenizer = LlamaTokenizer.from_pretrained(
    model_repo_id,
    use_auth_token='your_hf_access_token'  # 필요한 경우 액세스 토큰 추가
)

# 모델 로드
model = LlamaForCausalLM.from_pretrained(
    model_repo_id,
    torch_dtype=torch.float16,  # 또는 torch.bfloat16
    device_map="auto",          # 가능한 경우 GPU에 자동 할당
    use_auth_token='your_hf_access_token'  # 필요한 경우 액세스 토큰 추가
)

def respond(
    message,
    history,
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # 프롬프트 생성
    prompt = system_message + "\n"
    for user_msg, bot_msg in history:
        prompt += f"User: {user_msg}\nAssistant: {bot_msg}\n"
    prompt += f"User: {message}\nAssistant:"

    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

    outputs = model.generate(
        **inputs,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
        do_sample=True,
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.eos_token_id,
    )

    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    response = response[len(prompt):].strip()

    history.append((message, response))

    return history

demo = gr.ChatInterface(
    fn=respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", 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)",
        ),
    ],
)

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