app.py

import os
import gradio as gr
from dotenv import load_dotenv
from langchain.vectorstores.faiss import FAISS
from langchain.embeddings import HuggingFaceBgeEmbeddings
from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import CharacterTextSplitter

# Load environment variables
load_dotenv()

# Load and process the PDF files
loader = PyPDFLoader("./new_papers/ALiBi.pdf")
documents = loader.load()

# Split the documents into chunks and embed them using HuggingFaceBgeEmbeddings
text_splitter = CharacterTextSplitter(chunk_size=100, chunk_overlap=0)
vdocuments = text_splitter.split_documents(documents)

# Extract the text from the Document objects
docs_text = [doc.text for doc in vdocuments]

model = "BAAI/bge-base-en-v1.5"
encode_kwargs = {
    "normalize_embeddings": True
}  # set True to compute cosine similarity
embeddings = HuggingFaceBgeEmbeddings(
    model_name=model, encode_kwargs=encode_kwargs, model_kwargs={"device": "cpu"}
)

# Create FAISS vector store for API embeddings
api_db = FAISS.from_texts(texts=docs_text, embedding=embeddings)

# Define the PDF retrieval function
def pdf_retrieval(query):
    # Run the query through the retriever
    response = api_db.similarity_search(query)
    return response

# Create Gradio interface for the API retriever
api_tool = gr.Interface(
    fn=pdf_retrieval,
    inputs=[gr.Textbox()],
    outputs=gr.Textbox(),
    live=True,
    title="API PDF Retrieval Tool",
    description="This tool indexes PDF documents and retrieves relevant answers based on a given query (HuggingFaceBgeEmbeddings).",
)

# Launch the Gradio interface
api_tool.launch()