app.py

import transformers
import streamlit as st
from annotated_text import annotated_text


@st.cache(allow_output_mutation=True, show_spinner=False)
def get_pipe():
    tokenizer = transformers.AutoTokenizer.from_pretrained("deepset/roberta-base-squad2")
    model = transformers.AutoModelForQuestionAnswering.from_pretrained("deepset/roberta-base-squad2")
    pipe = transformers.pipeline("question-answering", model=model, tokenizer=tokenizer)
    return pipe


def parse_context(context, prediction):
    parsed_context = []
    parsed_context.append(context[:prediction["start"]])
    parsed_context.append((prediction["answer"], "ANSWER", "#afa"))
    parsed_context.append(context[prediction["end"]:])
    return parsed_context


st.set_page_config(page_title="Question Answering")
st.title("Question Answering")
st.write("Enter context and a question and press 'Predict' to extract the answer from the context.")

default_context = "My name is Wolfgang and I live in Berlin."
default_question = "What is my name?"

context = st.text_area("Enter context here:", value=default_context)
question = st.text_input("Enter question here:", value=default_question)
submit = st.button('Predict')

with st.spinner("Loading model..."):
    pipe = get_pipe()

if (submit and len(context.strip()) > 0 and len(question.strip()) > 0) or \
   (len(context.strip()) > 0 and len(question.strip()) > 0):

    prediction = pipe(question, context)

    parsed_context = parse_context(context, prediction)

    st.header("Prediction:")
    annotated_text(*parsed_context)

    st.header('Raw values:')
    st.json(prediction)