app.py

from transformers import AutoTokenizer, AutoModelForCausalLM
from sentence_transformers import SentenceTransformer
from datasets import load_dataset
import faiss
import gradio as gr
from accelerate import Accelerator
import os
import torch

# 환경 변수에서 Hugging Face API 키 로드
hf_api_key = os.getenv('HF_API_KEY')

# 모델 및 토크나이저 설정
model_id = "microsoft/phi-2"
tokenizer = AutoTokenizer.from_pretrained(model_id, token=hf_api_key, trust_remote_code=True)

# 토크나이저에 패딩 토큰 설정
if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token  # EOS 토큰을 패딩 토큰으로 사용

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    token=hf_api_key,
    trust_remote_code=True,
    torch_dtype=torch.float32
)

accelerator = Accelerator()
model = accelerator.prepare(model)

ST = SentenceTransformer("mixedbread-ai/mxbai-embed-large-v1")
dataset = load_dataset("not-lain/wikipedia", revision="embedded")
data = dataset["train"]
data = data.add_faiss_index("embeddings")

def search(query: str, k: int = 3):
    embedded_query = ST.encode(query)
    scores, retrieved_examples = data.get_nearest_examples("embeddings", embedded_query, k=k)
    return scores, retrieved_examples

def format_prompt(prompt, retrieved_documents, k):
    PROMPT = f"Question:{prompt}\nContext:"
    for idx in range(k):
        PROMPT += f"{retrieved_documents['text'][idx]}\n"
    return PROMPT

def generate(formatted_prompt):
    prompt_text = f"{SYS_PROMPT} {formatted_prompt}"
    input_ids = tokenizer(prompt_text, return_tensors="pt", padding="max_length", max_length=512).input_ids.to(accelerator.device)
    outputs = model.generate(
        input_ids,
        max_new_tokens=1024,
        eos_token_id=tokenizer.eos_token_id,
        do_sample=True,
        temperature=0.6,
        top_p=0.9
    )
    return tokenizer.decode(outputs[0][input_ids.shape[-1]:], skip_special_tokens=True)

def rag_chatbot_interface(prompt: str, k: int = 2):
    scores, retrieved_documents = search(prompt, k)
    formatted_prompt = format_prompt(prompt, retrieved_documents, k)
    return generate(formatted_prompt)

SYS_PROMPT = "You are an assistant for answering questions. Provide a conversational answer."

iface = gr.Interface(
    fn=rag_chatbot_interface,
    inputs="text",
    outputs="text",
    title="Retrieval-Augmented Generation Chatbot",
    description="This chatbot provides more accurate answers by searching relevant documents and generating responses."
)

iface.launch(share=True)