app.py

import gradio as gr
import re
import numpy as np
from langchain_community.vectorstores import Chroma
from langchain_community.embeddings import HuggingFaceBgeEmbeddings
import os
HUGGINGFACEHUB_API_TOKEN = os.environ["token"]

def clean_(l):
    s = list(l)[0][1]
    s = s.replace("\n", "=")
    return re.split('=', s, maxsplit=1)[-1].strip()

def similarity_search2(vectordb, query, k, unique="True"):
    print(f"\nQuery Key: {query}, \nrows requested:{k}\nUnique values:{unique}")
    D = vectordb.similarity_search(query,k)
    temp = []
    for d in D:
        temp.append(clean_(d))
    del D
    if unique == "True":
        return str(np.unique(np.array(temp)))[1:-1]
    else:
        return str(np.array(temp))[1:-1]
    
with gr.Blocks() as demo:
    gr.Markdown(
    """
    <h2> <center> Query Retrieval </center> </h2>
    """)
    with gr.Row():
        with gr.Column():
            query = gr.Textbox(placeholder="your query", label="Query")
            k = gr.Slider(10,100000,5, label="number of samples to check")
            unique = gr.Radio(["True", "False"], label="Return Unique values")
            with gr.Row():
                btn = gr.Button("Submit")
            def mmt_query(query, k, unique):
                model_id = "BAAI/bge-large-en-v1.5"
                model_kwargs = {"device": "cpu"}
                embedding = HuggingFaceBgeEmbeddings(
                    model_name = model_id,
                    model_kwargs = model_kwargs,
                    encode_kwargs = {'normalize_embeddings':True}
                )
                persist_directory = "db_book_mmt"
                vectordb = Chroma(persist_directory=persist_directory, embedding_function=embedding)
                return similarity_search2(vectordb, query, k, unique)
        with gr.Column():
            output = gr.Textbox(scale=10, label="Output")
            btn.click(mmt_query, [query, k, unique], output)
            
# demo.queue()
demo.launch()