import streamlit as st
import os
from transformers import AutoModelForCausalLM, AutoTokenizer

st.title("Meta LLaMA Text Generation")

@st.cache_resource
def load_model():
    model_name = "meta-llama/Meta-Llama-3-8B"
    access_token = os.getenv('hf')
    
    tokenizer = AutoTokenizer.from_pretrained(model_name, use_auth_token=access_token)
    model = AutoModelForCausalLM.from_pretrained(model_name, use_auth_token=access_token)
    return tokenizer, model

tokenizer, model = load_model()

prompt = st.text_input("Enter a prompt:", "Once upon a time")

if st.button("Generate Text"):
    inputs = tokenizer(prompt, return_tensors="pt")
    outputs = model.generate(**inputs, max_length=50)
    generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
    st.write(generated_text)