Spaces:
Sleeping
Sleeping
| import os | |
| import gradio as gr | |
| import asyncio | |
| from datetime import datetime | |
| from langchain_core.prompts import PromptTemplate | |
| from langchain_community.document_loaders import PyPDFLoader | |
| from langchain_google_genai import ChatGoogleGenerativeAI | |
| import google.generativeai as genai | |
| from langchain.chains.question_answering import load_qa_chain # Import load_qa_chain | |
| # Initialize an empty list to store chat history and context | |
| chat_history = [] | |
| context_history = "" | |
| async def initialize(file_path, question): | |
| global context_history | |
| genai.configure(api_key=os.getenv("GOOGLE_API_KEY")) | |
| model = genai.GenerativeModel('gemini-pro') | |
| model = ChatGoogleGenerativeAI(model="gemini-pro", temperature=0.3) | |
| # Refined prompt template to encourage precise and concise answers | |
| prompt_template = """Answer the question precisely and concisely using the provided context. Avoid any additional commentary or system messages. | |
| If the answer is not contained in the context, respond with "answer not available in context". | |
| Context: | |
| {context} | |
| Question: | |
| {question} | |
| Answer: | |
| """ | |
| prompt = PromptTemplate(template=prompt_template, input_variables=["context", "question"]) | |
| if os.path.exists(file_path): | |
| pdf_loader = PyPDFLoader(file_path) | |
| pages = pdf_loader.load_and_split() | |
| # Extract content from each page and store along with page number | |
| page_contexts = [page.page_content for i, page in enumerate(pages)] | |
| context = "\n".join(page_contexts[:30]) # Using the first 30 pages for context | |
| # Load the question-answering chain | |
| stuff_chain = load_qa_chain(model, chain_type="stuff", prompt=prompt) | |
| # Combine previous context with the new context | |
| combined_context = context_history + "\n" + context | |
| # Get the answer from the model | |
| stuff_answer = await stuff_chain.ainvoke({"input_documents": pages, "question": question, "context": combined_context}) | |
| answer = stuff_answer.get('output_text', '').strip() | |
| # Identify key sentences or phrases | |
| key_phrases = answer.split(". ") # Split answer into sentences for more precise matching | |
| # Score each page based on the presence of key phrases | |
| page_scores = [0] * len(pages) | |
| for i, page in enumerate(pages): | |
| for phrase in key_phrases: | |
| if phrase.lower() in page.page_content.lower(): | |
| page_scores[i] += 1 | |
| # Determine the top pages based on highest scores | |
| top_pages_with_scores = sorted(enumerate(page_scores), key=lambda x: x[1], reverse=True) | |
| top_pages = [i + 1 for i, score in top_pages_with_scores if score > 0][:2] # Get top 2 pages | |
| # Generate links for each top page | |
| file_name = os.path.basename(file_path) | |
| page_links = [f"[Page {p}](file://{os.path.abspath(file_path)})" for p in top_pages] | |
| page_links_str = ', '.join(page_links) | |
| if top_pages: | |
| source_str = f"Top relevant page(s): {page_links_str}" | |
| else: | |
| source_str = "Top relevant page(s): Not found in specific page" | |
| # Create a clickable link for the document | |
| source_link = f"[Document: {file_name}](file://{os.path.abspath(file_path)})" | |
| # Save interaction to chat history | |
| timestamp = datetime.now().isoformat() | |
| chat_history.append({ | |
| 'timestamp': timestamp, | |
| 'question': question, | |
| 'answer': answer, | |
| 'source': source_str, | |
| 'document_link': source_link | |
| }) | |
| # Update context history | |
| context_history += f"\nQ: {question}\nA: {answer}" | |
| return f"Answer: {answer}\n{source_str}\n{source_link}" | |
| else: | |
| return "Error: Unable to process the document. Please ensure the PDF file is valid." | |
| # Define Gradio Interface for QA and Chat History | |
| input_file = gr.File(label="Upload PDF File") | |
| input_question = gr.Textbox(label="Ask about the document") | |
| output_text = gr.Textbox(label="Answer and Top Pages", lines=10, max_lines=10) | |
| def get_chat_history(): | |
| history_str = "\n".join([f"Q: {entry['question']}\nA: {entry['answer']}\n{entry['source']}\n{entry['document_link']}\nTimestamp: {entry['timestamp']}\n" for entry in chat_history]) | |
| return history_str | |
| async def pdf_qa(file, question): | |
| if file is None: | |
| return "Error: No file uploaded. Please upload a PDF document." | |
| answer = await initialize(file.name, question) | |
| return answer | |
| # Create Gradio Interfaces | |
| qa_interface = gr.Interface( | |
| fn=pdf_qa, | |
| inputs=[input_file, input_question], | |
| outputs=output_text, | |
| title="PDF Question Answering System", | |
| description="Upload a PDF file and ask questions about the content." | |
| ) | |
| history_interface = gr.Interface( | |
| fn=get_chat_history, | |
| inputs=[], | |
| outputs=gr.Textbox(label="Chat History", lines=20, max_lines=20), | |
| title="Chat History", | |
| description="View the history of interactions." | |
| ) | |
| # Launch both interfaces | |
| qa_interface.launch(share=True) | |
| history_interface.launch(share=True) | |