app.py

import os
import torch
from accelerate import Accelerator
from PIL import Image
import random
import requests
import streamlit as st
from transformers import BlipProcessor, BlipForConditionalGeneration
from langchain_huggingface import HuggingFaceEndpoint
from langchain_core.prompts import PromptTemplate
from langchain_core.output_parsers import StrOutputParser

# Define the model IDs
llm_model_id = "mistralai/Mistral-7B-Instruct-v0.3"
blip_model_id = "Salesforce/blip-image-captioning-large"

# Initialize BLIP processor and model
processor = BlipProcessor.from_pretrained(blip_model_id)
model = BlipForConditionalGeneration.from_pretrained(blip_model_id)

# Initialize the accelerator
accelerator = Accelerator()

def get_llm_hf_inference(model_id=llm_model_id, max_new_tokens=128, temperature=0.1):
    try:
        llm = HuggingFaceEndpoint(
            repo_id=model_id,
            max_new_tokens=max_new_tokens,
            temperature=temperature,
            token=os.getenv("HF_TOKEN")
        )
    except Exception as e:
        st.error(f"Error loading model: {e}")
        llm = None
    return llm

def generate_caption(image, min_len=30, max_len=100):
    try:
        inputs = processor(image, return_tensors="pt")
        out = model.generate(**inputs, min_length=min_len, max_length=max_len)
        caption = processor.decode(out[0], skip_special_tokens=True)
        return caption
    except Exception as e:
        st.error(f"Error generating caption: {e}")
        return 'Unable to generate caption.'

# Configure the Streamlit app
st.set_page_config(page_title="HuggingFace ChatBot", page_icon="🤗")
st.title("Personal HuggingFace ChatBot")
st.markdown(f"*This is a simple chatbot using the HuggingFace transformers library with {llm_model_id}.*")

# Initialize session state
if "avatars" not in st.session_state:
    st.session_state.avatars = {'user': None, 'assistant': None}

if 'user_text' not in st.session_state:
    st.session_state.user_text = None

if "max_response_length" not in st.session_state:
    st.session_state.max_response_length = 256

if "system_message" not in st.session_state:
    st.session_state.system_message = "friendly AI conversing with a human user"

if "starter_message" not in st.session_state:
    st.session_state.starter_message = "Hello, there! How can I help you today?"

if "uploaded_image_path" not in st.session_state:
    st.session_state.uploaded_image_path = None

# Sidebar for settings
with st.sidebar:
    st.header("System Settings")
    st.session_state.system_message = st.text_area(
        "System Message", value="You are a friendly AI conversing with a human user."
    )
    st.session_state.starter_message = st.text_area(
        'First AI Message', value="Hello, there! How can I help you today?"
    )
    st.session_state.max_response_length = st.number_input(
        "Max Response Length", value=128
    )
    st.markdown("*Select Avatars:*")
    col1, col2 = st.columns(2)
    with col1:
        st.session_state.avatars['assistant'] = st.selectbox(
            "AI Avatar", options=["🤗", "💬", "🤖"], index=0
        )
    with col2:
        st.session_state.avatars['user'] = st.selectbox(
            "User Avatar", options=["👤", "👱‍♂️", "👨🏾", "👩", "👧🏾"], index=0
        )
    reset_history = st.button("Reset Chat History")

# Initialize or reset chat history
if "chat_history" not in st.session_state or reset_history:
    st.session_state.chat_history = [{"role": "assistant", "content": st.session_state.starter_message}]

def get_response(system_message, chat_history, user_text, max_new_tokens=256):
    hf = get_llm_hf_inference(max_new_tokens=max_new_tokens, temperature=0.1)
    if hf is None:
        return "Error with model inference.", chat_history

    prompt = PromptTemplate.from_template(
        "[INST] {system_message}\nCurrent Conversation:\n{chat_history}\n\nUser: {user_text}.\n [/INST]\nAI:"
    )
    chat = prompt | hf.bind(skip_prompt=True) | StrOutputParser(output_key='content')
    response = chat.invoke(input=dict(system_message=system_message, user_text=user_text, chat_history=chat_history))
    response = response.split("AI:")[-1]

    chat_history.append({'role': 'user', 'content': user_text})
    chat_history.append({'role': 'assistant', 'content': response})
    return response, chat_history

# Chat interface
chat_interface = st.container()
with chat_interface:
    output_container = st.container()
    
    # Image upload and captioning
    uploaded_image = st.file_uploader("Upload an image", type=["jpg", "jpeg", "png"])
    if uploaded_image and st.session_state.uploaded_image_path is None:
        # Save the uploaded image to a session-local directory
        with st.spinner("Processing image... 0%"):
            image = Image.open(uploaded_image).convert("RGB")
            
            # Create a directory for session images if not exists
            if not os.path.exists("session_images"):
                os.makedirs("session_images")
                
            # Save image to local session directory
            image_path = os.path.join("session_images", uploaded_image.name)
            image.save(image_path)
            
            # Generate and save caption
            caption = generate_caption(image)
            st.session_state.chat_history.append({'role': 'user', 'content': f'![uploaded image]({image_path})'})
            st.session_state.chat_history.append({'role': 'assistant', 'content': caption})
            st.spinner("Processing image... 100%")

    st.session_state.user_text = st.chat_input(placeholder="Enter your text here.")
    
    if st.session_state.user_text:
        with st.chat_message("user", avatar=st.session_state.avatars['user']):
            st.markdown(st.session_state.user_text)
        with st.chat_message("assistant", avatar=st.session_state.avatars['assistant']):
            response, st.session_state.chat_history = get_response(
                system_message=st.session_state.system_message,
                chat_history=st.session_state.chat_history,
                user_text=st.session_state.user_text,
                max_new_tokens=st.session_state.max_response_length
            )
            st.markdown(response)
            st.spinner("Thinking... 100%")