clqna.py

import streamlit as st

from transformers import pipeline
from pinecone import Pinecone, ServerlessSpec
from sentence_transformers import SentenceTransformer, util


bi_encoder = SentenceTransformer('msmarco-distilbert-base-v4')
bi_encoder.max_seq_length = 256     # Truncate long documents to 256 tokens

# Store the index as a variable
INDEX_NAME = 'cl-search-idx'
NAMESPACE = 'webpages'

index = pc.Index(name=INDEX_NAME)

def query_from_pinecone(index, question_embedding, top_k=3):
    # get embedding from THE SAME embedder as the documents

    return index.query(
      vector=question_embedding,
      top_k=top_k,
      namespace=NAMESPACE,
      include_metadata=True   # gets the metadata (dates, text, etc)
    ).get('matches')


QUESTION=st.text_area('Ask a question -e.g How to prepare for Verbal section for CAT?')  ##' How to prepare for Verbal section ?'
question_embedding = bi_encoder.encode(QUESTION, convert_to_tensor=True)
resp= query_from_pinecone(question_embedding.tolist(), 3)
docresult= resp[0]['metadata']['text'] 
#+ '\n*************\n'+  resp[1]['metadata']['text'] + '\n*************\n'+ resp[2]['metadata']['text']

st.json(out)