Spaces:
Runtime error
Runtime error
| from langchain_community.document_loaders import PyPDFLoader | |
| from langchain_text_splitters import RecursiveCharacterTextSplitter | |
| from langchain_openai import OpenAIEmbeddings | |
| from langchain_community.vectorstores import FAISS | |
| from langchain_openai import ChatOpenAI | |
| from langchain.retrievers import ContextualCompressionRetriever | |
| from langchain.retrievers.document_compressors import LLMChainExtractor | |
| from langchain.tools.retriever import create_retriever_tool | |
| from langchain import hub | |
| from langchain.agents import AgentExecutor, create_openai_tools_agent | |
| import os | |
| import gradio as gr | |
| # The Agent retriever is based on: https://python.langchain.com/docs/use_cases/question_answering/conversational_retrieval_agents?ref=blog.langchain.dev | |
| # The chat history is based on: https://python.langchain.com/docs/use_cases/question_answering/chat_history | |
| # Inspired by https://github.com/Niez-Gharbi/PDF-RAG-with-Llama2-and-Gradio/tree/master | |
| # Inspired by https://github.com/mirabdullahyaser/Retrieval-Augmented-Generation-Engine-with-LangChain-and-Streamlit/tree/master | |
| class PDFChatBot: | |
| # Initialize the class with the api_key and the model_name | |
| def __init__(self, api_key): | |
| self.processed = False | |
| self.final_agent = None | |
| self.chat_history = [] | |
| self.api_key = api_key | |
| self.llm = ChatOpenAI(openai_api_key=self.api_key, temperature=0, model_name="gpt-3.5-turbo-0125") | |
| # add text to Gradio text block (not needed without Gradio) | |
| def add_text(self, history, text): | |
| if not text: | |
| raise gr.Error("Please enter text.") | |
| history.append((text, '')) | |
| return history | |
| # Load a pdf document with langchain textloader | |
| def load_document(self, file_name): | |
| loader = PyPDFLoader(file_name) | |
| raw_document = loader.load() | |
| return raw_document | |
| # Split the document | |
| def split_documents(self, raw_document): | |
| text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, | |
| chunk_overlap=100, | |
| length_function=len, | |
| is_separator_regex=False, | |
| separators=["\n\n", "\n", " ", ""]) | |
| chunks = text_splitter.split_documents(raw_document) | |
| return chunks | |
| # Embed the document with OpenAI Embeddings & store it to vectorstore | |
| def create_retriever(self, chunks): | |
| embedding_func = OpenAIEmbeddings(openai_api_key=self.api_key) | |
| # Create a new vectorstore from the chunks | |
| vectorstore = FAISS.from_documents(chunks, embedding_func) | |
| # Create a retriever | |
| basic_retriever = vectorstore.as_retriever() | |
| compressor = LLMChainExtractor.from_llm(self.llm) | |
| compression_retriever = ContextualCompressionRetriever(base_compressor=compressor, | |
| base_retriever=basic_retriever) | |
| return basic_retriever # or compression_retriever | |
| # Create an agent | |
| def create_agent(self, retriever): | |
| tool = create_retriever_tool(retriever, | |
| f"search_document", | |
| f"Searches and returns excerpts from the provided document.") | |
| tools = [tool] | |
| prompt = hub.pull("hwchase17/openai-tools-agent") | |
| agent = create_openai_tools_agent(self.llm, tools, prompt) | |
| self.final_agent = AgentExecutor(agent=agent, tools=tools) | |
| # Process files | |
| def process_file(self, file_name): | |
| documents = self.load_document(file_name) | |
| texts = self.split_documents(documents) | |
| db = self.create_retriever(texts) | |
| self.create_agent(db) | |
| print("Files successfully processed") | |
| # Generate a response and write to memory | |
| def generate_response(self, history, query, path): | |
| if not self.processed: | |
| self.process_file(path) | |
| self.processed = True | |
| result = self.final_agent.invoke({'input': query, 'chat_history': self.chat_history})['output'] | |
| self.chat_history.extend((query, result)) | |
| for char in result: # history argument and the subsequent code is only for the purpose of Gradio | |
| history[-1][1] += char | |
| return history, " " | |