app.py

import torch
from PIL import Image
from transformers import AutoModel, AutoTokenizer
import streamlit as st
from transformers import pipeline
from huggingface_hub import InferenceClient
import os

# Define your API key here
my_key = "your_api_key_here"

# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained('openbmb/MiniCPM-V-2', trust_remote_code=True)
model = AutoModel.from_pretrained('openbmb/MiniCPM-V-2',trust_remote_code=True)
model.eval()

# Set device for model
device = 'cuda' if torch.cuda.is_available() else 'cpu'
model = model.to(device=device, dtype=torch.float16 if device == 'cuda' else torch.float32)

# Retrieve the API key from the environment
api_key = os.getenv("HF_API_KEY")

# Initialize the Hugging Face Inference client with the API key
client = InferenceClient(api_key=api_key)

# Streamlit UI setup
st.title("Image Questioning and Content Generation App")
st.write("Upload an image and ask a question. The model will respond with a description, and you can generate a song or story based on the response.")

# Upload an image
uploaded_image = st.file_uploader("Upload an image", type=["jpg", "jpeg", "png"])
if uploaded_image:
    image = Image.open(uploaded_image).convert('RGB')
    st.image(image, caption="Uploaded Image", use_column_width=True)

# Text input for the question
question = st.text_input("Ask a question about the image")
if question and uploaded_image:
    msgs = [{'role': 'user', 'content': question}]
    
    # Model's response to the image question
    with st.spinner("Processing..."):
        res, context, _ = model.chat(
            image=image,
            msgs=msgs,
            context=None,
            tokenizer=tokenizer,
            sampling=True,
            temperature=0.7
        )
    
    st.write("Model's response:", res)

    # Options for generating content based on the response
    option = st.selectbox("Generate content based on the response", ["Choose...", "Write a Song", "Write a Story"])

    if option != "Choose...":
        # Create a message based on user choice
        if option == "Write a Song":
            messages = [{"role": "user", "content": f"Write a song about the following: {res}"}]
        elif option == "Write a Story":
            messages = [{"role": "user", "content": f"Write a story about the following: {res}"}]

        # Stream the content generation
        st.write(f"Generating {option.lower()}...")

        stream = client.chat.completions.create(
            model="meta-llama/Llama-3.2-3B-Instruct", 
            messages=messages, 
            max_tokens=500,
            stream=True
        )

        generated_text = ""
        for chunk in stream:
            generated_text += chunk.choices[0].delta.content
            st.write(generated_text)  # Display each chunk as it's generated