import os from langchain_community.document_loaders import PyPDFLoader from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain_huggingface.embeddings import HuggingFaceEmbeddings from langchain_community.vectorstores import FAISS from langchain_huggingface import HuggingFaceEndpoint from langchain.prompts import PromptTemplate from langchain.schema.runnable import RunnablePassthrough from langchain.schema.output_parser import StrOutputParser import gradio as gr ############################## OLD CODE FOR 1 PDF #################################### # Load and split documents FILE_PATH = "" loader = PyPDFLoader(FILE_PATH) docs = loader.load() text_splitter = RecursiveCharacterTextSplitter( chunk_size=1000, chunk_overlap=200, ) splits = text_splitter.split_documents(docs) # Set up embeddings and vector store HF_EMBED_MODEL_ID = "BAAI/bge-small-en-v1.5" embeddings = HuggingFaceEmbeddings(model_name=HF_EMBED_MODEL_ID) vectorstore = FAISS.from_documents(splits, embeddings) ############################## NEW CODE FOR 5 PDFs #################################### # Load and split documents # FILE_PATHS = ["vol1.pdf", "vol2.pdf"] # # Initialize an empty list to store all documents # all_docs = [] # # Iterate through each file path in FILE_PATHS # for file_path in FILE_PATHS: # # Load the PDF using PyPDFLoader # loader = PyPDFLoader(file_path) # # Split the loaded document using RecursiveCharacterTextSplitter # text_splitter = RecursiveCharacterTextSplitter( # chunk_size=1000, # chunk_overlap=200, # ) # # Load and split the current document and append to all_docs # documents = loader.load_and_split(text_splitter) # all_docs.extend(documents) # # Set up embeddings and vector store # HF_EMBED_MODEL_ID = "BAAI/bge-small-en-v1.5" # embeddings = HuggingFaceEmbeddings(model_name=HF_EMBED_MODEL_ID) # vectorstore = FAISS.from_documents(all_docs, embeddings) # Set up LLM HF_API_KEY = os.environ.get("HF_API_KEY") # Ensure this is set in Hugging Face Space settings HF_LLM_MODEL_ID = "mistralai/Mistral-7B-Instruct-v0.3" llm = HuggingFaceEndpoint( repo_id=HF_LLM_MODEL_ID, huggingfacehub_api_token=HF_API_KEY, ) # Create RAG pipeline retriever = vectorstore.as_retriever() prompt = PromptTemplate.from_template( "Context information is below.\\n---------------------\\n{context}\\n---------------------\\nGiven the context information and not prior knowledge, answer the query.\\nQuery: {question}\\nAnswer:\\n" ) rag_chain = ( {"context": retriever, "question": RunnablePassthrough()} | prompt | llm | StrOutputParser() ) # Gradio interface def ask_question(question): return rag_chain.invoke(question) iface = gr.Interface(fn=ask_question, inputs="text", outputs="text") iface.launch()