""" File: RAG-Demo.py Author: Martin Rattensberger Description: A GUI application for interacting with a local Llama vision model. Users can select a directory with PDF files and ask questions about them. Date: 11.11.2024 # Replace with actual date Version: 1.1 Development Environment: Visual Studio Code with Continue.ai (Claude Sonnet 3.5) This script creates a tkinter-based GUI for selecting a directory with PDFs, sending them to a local Llama 3.2 vision model, and displaying the results. """ import tkinter as tk from tkinter import filedialog, scrolledtext import ollama from PIL import Image import fitz # PyMuPDF library for handling PDFs import io import base64 import threading import time import os class LlamaVisionApp: def __init__(self, master): self.master = master master.title("Llama Vision Interface") # Directory selection button self.select_dir_button = tk.Button(master, text="Select PDF Directory", command=self.select_directory) self.select_dir_button.pack(pady=10) # Display selected directory self.dir_label = tk.Label(master, text="No directory selected") self.dir_label.pack() # PDF file listbox self.pdf_listbox = tk.Listbox(master, width=50, height=5) self.pdf_listbox.pack(pady=10) # Question input self.question_entry = tk.Text(master, width=50, height=3) self.question_entry.pack(pady=10) self.question_entry.insert(tk.END, "What is in this PDF?") # Submit button self.submit_button = tk.Button(master, text="Submit", command=self.submit_question) self.submit_button.pack() # Response display self.response_text = scrolledtext.ScrolledText(master, width=60, height=30) self.response_text.pack(pady=10) self.directory_path = None self.pdf_files = [] self.image_data = None self.processing = False def select_directory(self): self.directory_path = filedialog.askdirectory() if self.directory_path: self.dir_label.config(text=f"Selected directory: {self.directory_path}") self.load_pdf_files() def load_pdf_files(self): self.pdf_files = [f for f in os.listdir(self.directory_path) if f.lower().endswith('.pdf')] self.pdf_listbox.delete(0, tk.END) for pdf in self.pdf_files: self.pdf_listbox.insert(tk.END, pdf) def load_selected_pdf(self): selected_indices = self.pdf_listbox.curselection() if not selected_indices: return None selected_pdf = self.pdf_files[selected_indices[0]] pdf_path = os.path.join(self.directory_path, selected_pdf) pdf_document = fitz.open(pdf_path) first_page = pdf_document[0] image = first_page.get_pixmap() img = Image.frombytes("RGB", [image.width, image.height], image.samples) buffer = io.BytesIO() img.save(buffer, format="PNG") image_data = base64.b64encode(buffer.getvalue()).decode('utf-8') pdf_document.close() return image_data def submit_question(self): self.image_data = self.load_selected_pdf() if not self.image_data: self.response_text.delete('1.0', tk.END) self.response_text.insert(tk.END, "Please select a PDF file first.\n") return question = self.question_entry.get('1.0', tk.END).strip() self.response_text.delete('1.0', tk.END) self.processing = True threading.Thread(target=self.processing_animation).start() threading.Thread(target=self.run_llama_model, args=(question,)).start() def processing_animation(self): animation = "|/-\\" i = 0 while self.processing: self.response_text.delete('1.0', tk.END) self.response_text.insert(tk.END, f"Processing {animation[i % len(animation)]}") self.master.update_idletasks() time.sleep(0.1) i += 1 def run_llama_model(self, question): try: response = ollama.chat( model='llama3.2-vision', messages=[{ 'role': 'user', 'content': question, 'images': [self.image_data] }] ) self.processing = False self.master.after(0, self.update_response, question, response['message']['content']) except Exception as e: self.processing = False self.master.after(0, self.update_response, question, f"Error: {str(e)}") def update_response(self, question, answer): self.response_text.delete('1.0', tk.END) self.response_text.insert(tk.END, f"Q: {question}\nA: {answer}\n\n") root = tk.Tk() app = LlamaVisionApp(root) root.mainloop()