Spaces:
Sleeping
Sleeping
| import os | |
| import gradio as gr | |
| import asyncio | |
| from langchain_core.prompts import PromptTemplate | |
| from langchain_community.output_parsers.rail_parser import GuardrailsOutputParser | |
| 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 | |
| async def initialize(file_path, question): | |
| 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() | |
| context = "\n".join(f"Page {i+1}: {page.page_content}" for i, page in enumerate(pages[:30])) | |
| stuff_chain = load_qa_chain(model, chain_type="stuff", prompt=prompt) | |
| # Use ainvoke to get the result | |
| stuff_answer = await stuff_chain.ainvoke({"input_documents": pages, "question": question, "context": context}) | |
| # Access the correct key for the answer | |
| answer = stuff_answer.get('output_text', '').strip() | |
| # Extract the page number where the context was found | |
| sources = [] | |
| for i, page in enumerate(pages): | |
| if question.lower() in page.page_content.lower(): | |
| sources.append(f"Page {i+1}") | |
| if sources: | |
| source_str = f" (Source: {', '.join(sources)})" | |
| else: | |
| source_str = " (Source: Not found in specific page)" | |
| # Add the clickable link to the source | |
| file_name = os.path.basename(file_path) | |
| source_link = f"[{file_name}](file://{os.path.abspath(file_path)})" | |
| return f"{answer} {source_str} - [Document: {source_link}]" | |
| else: | |
| return "Error: Unable to process the document. Please ensure the PDF file is valid." | |
| # Define Gradio Interface | |
| input_file = gr.File(label="Upload PDF File") | |
| input_question = gr.Textbox(label="Ask about the document") | |
| output_text = gr.Textbox(label="Answer - GeminiPro") | |
| async def pdf_qa(file, question): | |
| answer = await initialize(file.name, question) | |
| return answer | |
| # Create Gradio Interface with share=True to enable a public link | |
| 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.").launch(share=True) | |