app.py

import streamlit as st
import json 
from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline
from sentence_transformers import SentenceTransformer, util
from io import StringIO

@st.cache
def vectorizeSentences(model, sentences):
    embeddings = model.encode(sentences)
    cosine_scores = util.cos_sim(embeddings[0], embeddings)
    return cosine_scores[0]

@st.cache
def loadContext(model, context):
    data = json.load(contextUpload)
    embeddings = vectorizeSentences(model, list(data.keys()))
    return (data, embeddings)

@st.cache
def question(model, question, data, embeddings):
    data = json.load(contextUpload)
    embeddings = vectorizeSentences(model, list(data.keys()))
    return (data, embeddings)


@st.cache
def loadSentenceModel():
    return SentenceTransformer('sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2')

@st.cache
def loadQAModel():
    return pipeline("question-answering", model="deepset/roberta-base-squad2", tokenizer="deepset/roberta-base-squad2")

sentenceModel = loadSentenceModel
questionAnsweringModel = loadQAModel

st.header("Large Context Question & Answering")

st.info("Upload a JSON context file")

contextUpload = st.file_uploader("Upload a.json context file", type=["json"])
st.json(json.load(contextUpload))
if contextUpload is not None:
    #embeddings = loadContext(sentenceModel, contextUpload)
   # question = st.text_input("Enter your question", value="")
    #if question != "":
    st.write("question")