app.py

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

# Load models and tokenizers
@st.cache_resource
def load_model_and_tokenizer(model_name):
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    # Load the model in 8-bit quantization
    model = AutoModelForCausalLM.from_pretrained(model_name, device_map="cpu", load_in_8bit=True)
    return model, tokenizer

model_8b, tokenizer_8b = load_model_and_tokenizer("huggyllama/llama-3.1-8b")
model_8b_instruct, tokenizer_8b_instruct = load_model_and_tokenizer("huggyllama/llama-3.1-8b-instruct")

def generate_text(model, tokenizer, prompt, max_length=100):
    inputs = tokenizer(prompt, return_tensors="pt")
    with torch.no_grad():
        outputs = model.generate(**inputs, max_length=max_length, num_return_sequences=1)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

st.title("LLaMA-3.1-8B vs LLaMA-3.1-8B-Instruct Comparison (CPU Version)")

prompt = st.text_area("Enter your prompt:", height=100)
max_length = st.slider("Max output length:", min_value=50, max_value=500, value=100)

if st.button("Generate"):
    if prompt:
        st.warning("Generation may take several minutes. Please be patient.")
        
        col1, col2 = st.columns(2)
        
        with col1:
            st.subheader("LLaMA-3.1-8B Output")
            output_8b = generate_text(model_8b, tokenizer_8b, prompt, max_length)
            st.write(output_8b)
        
        with col2:
            st.subheader("LLaMA-3.1-8B-Instruct Output")
            output_8b_instruct = generate_text(model_8b_instruct, tokenizer_8b_instruct, prompt, max_length)
            st.write(output_8b_instruct)
    else:
        st.warning("Please enter a prompt.")