app.py

import streamlit as st
from transformers import pipeline
import torch

# Set device
device = 0 if torch.cuda.is_available() else -1

# Initialize the pipeline with reduced precision
@st.cache_resource
def load_pipeline():
    return pipeline(
        "text-generation",
        model="Irina-Igmm/llama2-Immo",
        torch_dtype=torch.float16,  # Use float16 for reduced memory usage
        device=device,  # Use GPU if available
    )

pipe = load_pipeline()

st.title("Demande informations avec Llama2-Immo")

# User input
user_input = st.text_area("Entrer votre invite:", "")

if st.button("Generate"):
    if user_input:
        # Process the input
        with st.spinner("Generating..."):
            result = pipe(user_input, max_length=500)  # Adjust max_length as needed
        # Display the output
        st.subheader("Informations demandées")
        st.write(result[0]["generated_text"])
    else:
        st.warning("S'il vous plaît, entrez un prompt.")

# Clean up
torch.cuda.empty_cache()