import gradio as gr import PyPDF2 import io import time import os from together import Together def extract_text_from_pdf(pdf_file): text="" try: if hasattr(pdf_file, 'read'): pdf_content = pdf_file.read() if hasattr(pdf_file, 'seek'): pdf_file.seek(0) else: pdf_content = pdf_file # Read the PDF file pdf_reader = PyPDF2.PdfReader(io.BytesIO(pdf_content)) #convert to files for page_num in range(len(pdf_reader.pages)): page_text = pdf_reader.pages[page_num].extract_text() if page_text: # Check if text extraction worked text += page_text + "\n\n" else: text += f"[Page {page_num+1} - No extractable text found]\n\n" if not text.strip(): return "No text could be extracted from the PDF. The document may be scanned or image-based." return text except Exception as e: return f"Error extracting text from PDF: {str(e)}" def chat_with_pdf(api_key, pdf_text, user_question, history): if not api_key.strip(): return history + [(user_question, "Error: Please enter your Together API key.")], history if not pdf_text.strip() or pdf_text.startswith("Error") or pdf_text.startswith("No text"): return history + [(user_question, "Error: Please upload a valid PDF file with extractable text first.")], history if not user_question.strip(): return history + [(user_question, "Error: Please enter a question.")], history try: client=Together(api_key=api_key) max_context_length=10000 if len(pdf_text) > max_context_length: # More sophisticated truncation that preserves beginning and end half_length = max_context_length // 2 ####20000/2 = 10000 pdf_context = pdf_text[:half_length] + "\n\n[...Content truncated due to length...]\n\n" + pdf_text[-half_length:] #0:5000 (first 5000) 20000:5000 (last 5000) else: pdf_context = pdf_text system_message = f"""You are an intelligent assistant designed to read, understand, and extract information from PDF documents. Based on any question or query the user asks—whether it's about content, summaries, data extraction, definitions, insights, or interpretation—you will analyze the following PDF content and provide an accurate, helpful response grounded in the document. Always respond with clear, concise, and context-aware information. PDF CONTENT: {pdf_context} Answer the user's questions only based on the PDF content above. If the answer cannot be found in the PDF, politely state that the information is not available in the provided document.""" messages = [ {"role": "system", "content": system_message}, ] # Add chat history for h_user, h_bot in history: messages.append({"role": "user", "content": h_user}) #conv1, ..... messages.append({"role": "assistant", "content": h_bot}) #conv1res, .... messages.append({"role": "user", "content": user_question}) #convx ----- History (context) response=client.chat.completions.create( model="meta-llama/Llama-3.3-70B-Instruct-Turbo-Free", messages=messages, max_tokens=5000, #5000 temperature=0.7, ) # Extract the assistant's response assistant_response = response.choices[0].message.content ## convx? Ans: convxres (error: ans (something is wrong: model out of limit)) new_history = history + [(user_question, assistant_response)] return new_history, new_history except Exception as e: error_message = f"Error: {str(e)}" return history + [(user_question, error_message)], history ### history + current question + error message def process_pdf(pdf_file, api_key_input): if pdf_file is None: return "Please upload a PDF file.", "", [] try: # Get the file name file_name = os.path.basename(pdf_file.name) if hasattr(pdf_file, 'name') else "Upload PDF" # Extract text from the PDF pdf_text = extract_text_from_pdf(pdf_file) # Check if there was an error in extraction if pdf_text.startswith("Error extracting text from PDF"): return f"❌ {pdf_text}", "", [] if not pdf_text.strip() or pdf_text.startswith("No text could be extracted"): return f"⚠️ {pdf_text}", "", [] # Count words for information word_count = len(pdf_text.split()) # Return a message with the file name and text content status_message = f"✅ Successfully processed PDF: {file_name} ({word_count} words extracted)" return status_message, pdf_text, [] except Exception as e: return f"❌ Error processing PDF: {str(e)}", "", [] def validate_api_key(api_key): """Simple validation for API key format""" if not api_key or not api_key.strip(): return "❌ API Key is required" if len(api_key.strip()) < 10: return "❌ API Key appears to be too short" return "✓ API Key format looks valid (not verified with server)" ###UI with gr.Blocks(title="ChatPDF with Together AI", theme=gr.themes.Ocean()) as app: gr.Markdown("# 📄 ChatPDF with Together AI") gr.Markdown("Upload a PDF and chat with it using the Llama-3.3-70B model.") with gr.Row(): with gr.Column(scale=1): # API Key input api_key_input = gr.Textbox( label="Together API Key", placeholder="Enter your Together API key here...", type="password" ) # API key validation api_key_status = gr.Textbox( label="API Key Status", interactive=False ) # PDF upload pdf_file = gr.File( label="Upload PDF", file_types=[".pdf"], type="binary" # Ensure we get binary data ) # Process PDF button process_button = gr.Button("Process PDF") # Status message status_message = gr.Textbox( label="Status", interactive=False ) with gr.Accordion("PDF Content Preview", open=True): pdf_preview = gr.Textbox( label="Extracted Text Preview", interactive=False, max_lines=10, show_copy_button=True ) with gr.Column(scale=2): # Chat interface chatbot = gr.Chatbot( label="Chat with PDF", height=500, show_copy_button=True ) # Question input question = gr.Textbox( label="Ask a question about the PDF", placeholder="What is the main topic of this document?", lines=2 ) # Submit button submit_button = gr.Button("Submit Question") # Event handlers def update_preview(text): """Update the preview with the first few lines of the PDF text""" if not text or text.startswith("Error") or text.startswith("No text"): return text # Get the first ~500 characters for preview preview = text[:500] if len(text) > 500: preview += "...\n[Text truncated for preview. Full text will be used for chat.]" return preview # API key validation event api_key_input.change( fn=validate_api_key, inputs=[api_key_input], outputs=[api_key_status] ) process_button.click( fn=process_pdf, inputs=[pdf_file, api_key_input], outputs=[status_message, pdf_text, chatbot] ).then( fn=update_preview, inputs=[pdf_text], outputs=[pdf_preview] ) submit_button.click( fn=chat_with_pdf, inputs=[api_key_input, pdf_text, question, chatbot], outputs=[chatbot, chatbot] ).then( fn=lambda: "", outputs=question ) question.submit( fn=chat_with_pdf, inputs=[api_key_input, pdf_text, question, chatbot], outputs=[chatbot, chatbot] ).then( fn=lambda: "", outputs=question ) # Launch the app if __name__ == "__main__": app.launch(share=True)