app.py

import gradio as gr
from langchain.vectorstores import Chroma
from langchain.document_loaders import PyPDFLoader
from langchain.embeddings import HuggingFaceInstructEmbeddings

# Initialize the HuggingFaceInstructEmbeddings
hf = HuggingFaceInstructEmbeddings(
    model_name="gpt2",
    embed_instruction="Represent the document for retrieval: ",
    query_instruction="Represent the query for retrieval: "
)
# Add a padding token to the tokenizer
hf.tokenizer.add_special_tokens({'pad_token': '[PAD]'})

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

#loader = PyPDFLoader('./new_papers/', glob="./*.pdf")
documents = loader.load()

# Create a Chroma vector store from the PDF documents
db = Chroma.from_documents(documents, hf, collection_name="my-collection")

class PDFRetrievalTool:
    def __init__(self):
        self.retriever = db.as_retriever(search_kwargs={"k": 1})

    def __call__(self, query):
        # Run the query through the retriever
        response = self.retriever.run(query)
        return response['result']

# Create the Gradio interface using the PDFRetrievalTool
tool = gr.Interface(
    PDFRetrievalTool(),
    inputs=gr.Textbox(),
    outputs=gr.Textbox(),
    live=True,
    title="PDF Retrieval Tool",
    description="This tool indexes PDF documents and retrieves relevant answers based on a given query.",
)

# Launch the Gradio interface
tool.launch()