app.py

import streamlit as st
from optimum.nvidia.pipelines import pipeline
import torch
import os
import sys



# Retrieve the Hugging Face token from environment variables
hf_token = os.environ.get("HF_TOKEN")

if not hf_token:
    st.error("Hugging Face token not found. Please add your HF_TOKEN to the Space secrets.")
    st.stop()


@st.cache_resource
def load_pipeline(model_name):
    with st.spinner(f'Loading {model_name}... This may take several minutes.'):
        try:
            pipe = pipeline("text-generation", model=model_name,use_fp8=True)
        except Exception as e:
            st.error(f"An error occurred: {e}")
            st.stop()
    return pipe

pipe8 = load_pipeline("unsloth/Meta-Llama-3.1-8B-bnb-4bit")
pipe8instruct = load_pipeline("SanctumAI/Meta-Llama-3.1-8B-Instruct-GGUF")

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")

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:
        col1, col2 = st.columns(2)
        
        with col1:
            st.subheader("LLaMA-3.1-8B Output")
            output_8b = pipe8(prompt, max_length)
            st.write(output_8b[0]['generated_text'])
        
        with col2:
            st.subheader("LLaMA-3.1-8B-Instruct Output")
            output_8b_instruct = pipe8instruct(prompt, max_length)
            st.write(output_8b_instruct[0]['generated_text'])
    else:
        st.warning("Please enter a prompt.")