import json; import streamlit as st; import requests as req; from transformers import pipeline
WIKI_URL = 'https://en.wikipedia.org/w/api.php'; WIKI_BERT = "&titles=BERT_(language_model)"
WIKI_QUERY = "?format=json&action=query&prop=extracts&explaintext=1"; WIKI_METHOD = 'GET'
pipe_exqa = pipeline("question-answering") #, model="distilbert-base-cased-distilled-squad"
st.title('Question Answering example')
st.subheader('1. A simple question (extractive, closed domain)')
response = req.request(WIKI_METHOD, f'{WIKI_URL}{WIKI_QUERY}{WIKI_BERT}')
resp_json = json.loads(response.content.decode("utf-8"))
wiki_bert = resp_json['query']['pages']['62026514']['extract']
paragraph = wiki_bert
par_text = 'Paragraph used for QA (you can also edit, or copy/paste new content)'
written_passage = st.text_area(par_text, paragraph, height=250)
if written_passage:
    paragraph = written_passage
question = 'How many attention heads does Bert have?' # question = 'How many languages does bert understand?'
query_text = 'Question used for QA (you can also edit, and experiment with the answers)'
written_question = st.text_input(query_text, question)
if written_question:
    question = written_question
QA_URL = "https://api-inference.huggingface.co/models/deepset/roberta-base-squad2"; QA_METHOD = 'POST'
if st.button('Run QA inference (get answer prediction)'):
    if paragraph and question:
        inputs = {'question': question, 'context': paragraph}
        payload = json.dumps(inputs)
        prediction = req.request(QA_METHOD, QA_URL, data=payload)
        answer = json.loads(prediction.content.decode("utf-8"))
        # >>> answer structure: # { "answer": "over 70", "score": 0.240, "start": 35, "end": 62 }
        answer_dict = dict(answer) # st.write(answer_dict)
        print(answer_dict)
        if "answer" in answer_dict.keys():
            answer_span, answer_score = answer_dict["answer"], answer_dict["score"]
            st.write(f'Answer: **{answer_span}**')
            start_par, stop_para = max(0, answer_dict["start"]-86), min(answer_dict["end"]+90, len(paragraph))
            answer_context = paragraph[start_par:stop_para].replace(answer_span, f'**{answer_span}**')
            st.write(f'Answer context (and score): ... _{answer_context}_ ... (score: {format(answer_score, ".3f")})')
            st.write(f'Answer JSON: '); st.write(answer)
        else:
            try:
                qa_result = pipe_exqa(question=question, context=paragraph)
            except Exception as e:
                qa_result = str(e)

            if "answer" in qa_result.keys():
                answer_span, answer_score = qa_result["answer"], qa_result["score"]
                st.write(f'Answer: **{answer_span}**')
                start_par, stop_para = max(0, qa_result["start"]-86), min(qa_result["end"]+90, len(paragraph))
                answer_context = paragraph[start_par:stop_para].replace(answer_span, f'**{answer_span}**')
                st.write(f'Answer context (and score): ... _{answer_context}_ ... (score: {format(answer_score, ".3f")})')
            
            st.write(f'Answer JSON: '); st.write(qa_result)
    else:
        st.write('Write some passage of text and a question'); st.stop()
# x = st.slider('Select a value'); st.write(x, 'squared is', x * x)