app.py

import os 
#!pip install -q gradio langchain pypdf chromadb
import gradio as gr
from langchain.vectorstores import Chroma
from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import CharacterTextSplitter
from langchain.embeddings import HuggingFaceInferenceAPIEmbeddings

# Use Hugging Face Inference API embeddings
inference_api_key = os.environ['HF']
api_hf_embeddings = HuggingFaceInferenceAPIEmbeddings(
    api_key=inference_api_key,
    model_name="sentence-transformers/all-MiniLM-l6-v2"
)

# Load and process the PDF files
loader = PyPDFLoader("./new_papers/ALiBi.pdf")
documents = loader.load()
print("-----------")
print(documents[0])
print("-----------")


# Extract the embedding arrays from the PDF documents
embeddings = []
for doc in documents:
    embeddings.extend(doc['embeddings'])

# Create Chroma vector store for API embeddings
api_db = Chroma.from_texts(embeddings, api_hf_embeddings, collection_name="api-collection")
        

# 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
# 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 (HF Inference API Embeddings).",
)

# Launch the Gradio interface
#api_tool.launch()