app.py

import os
import streamlit as st
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
import torch
from diffusers import StableDiffusionPipeline

# Get Hugging Face Token from Environment Variable
HF_TOKEN = os.environ.get("stablediffussion")

# Ensure the token is set
if not HF_TOKEN:
    st.error("Hugging Face token is missing. Please set it in your environment variables.")
    st.stop()

# Load the text model
@st.cache_resource
def load_text_model():
    model_name = "meta-llama/Llama-2-7b-chat-hf"

    tokenizer = AutoTokenizer.from_pretrained(model_name, token=HF_TOKEN)
    model = AutoModelForCausalLM.from_pretrained(
        model_name,
        device_map="cpu",  # Run on CPU
        torch_dtype=torch.float32,  # Use float32 for CPU
        token=HF_TOKEN
    )
    
    return pipeline("text-generation", model=model, tokenizer=tokenizer, device=-1)  # Force CPU

# Load the image model (Stable Diffusion)
@st.cache_resource
def load_image_model():
    model_id = "runwayml/stable-diffusion-v1-5"
    
    pipe = StableDiffusionPipeline.from_pretrained(
        model_id, torch_dtype=torch.float32, token=HF_TOKEN
    )
    pipe.to("cpu")  # Use CPU instead of GPU
    
    return pipe

# Load Models
llama_pipe = load_text_model()
image_pipe = load_image_model()

# Streamlit UI
st.title("🚀 Blog & Image Generator")

topic = st.text_input("Enter a blog topic:", "The Future of AI in Healthcare")

if st.button("Generate Blog"):
    with st.spinner("Generating Blog..."):
        response = llama_pipe(f"Write a detailed blog on {topic}:", max_length=800, truncation=True)
        blog = response[0]['generated_text']
        
        st.subheader("Generated Blog")
        st.write(blog)

        # Extract first 3 lines for image keywords
        keywords = " ".join(blog.split("\n")[:3])

        # Generate Image
        with st.spinner("Generating Image..."):
            image = image_pipe(keywords).images[0]
            st.subheader("Generated Image")
            st.image(image, caption="AI-Generated Image", use_column_width=True)