app.py

import streamlit as st
import pysbd
from transformers import pipeline
from sentence_transformers import CrossEncoder
from  transformers  import  AutoTokenizer, AutoModelWithLMHead, pipeline

model_name = "MaRiOrOsSi/t5-base-finetuned-question-answering"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelWithLMHead.from_pretrained(model_name)

#from transformers import pipeline

#text2text_generator = pipeline("text2text-generation", model = "gpt2")

sentence_segmenter = pysbd.Segmenter(language='en',clean=False)
passage_retreival_model = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2')
qa_model = pipeline("question-answering",'a-ware/bart-squadv2')

def fetch_answers(question, document ):
    document_paragraphs = document.splitlines()
    query_paragraph_list = [(question, para) for para in document_paragraphs if len(para.strip()) > 0 ]
    
    scores = passage_retreival_model.predict(query_paragraph_list)
    top_5_indices = scores.argsort()[-5:]
    top_5_query_paragraph_list = [query_paragraph_list[i] for i in top_5_indices ]
    top_5_query_paragraph_list.reverse()
    
    top_5_query_paragraph_answer_list = ""
    count = 1
    for query, passage in top_5_query_paragraph_list:
     passage_sentences = sentence_segmenter.segment(passage)
     answer = qa_model(question = query, context = passage)['answer']
     evidence_sentence = ""
     for i in range(len(passage_sentences)):
         if answer.startswith('.') or answer.startswith(':'):
             answer = answer[1:].strip()
         if answer in passage_sentences[i]:
             evidence_sentence = evidence_sentence + " " + passage_sentences[i]
     
                  
     model_input = f"question: {query} context: {evidence_sentence}"
     #output_answer = text2text_generator(model_input)[0]['generated_text']
     encoded_input = tokenizer([model_input],
                               return_tensors='pt',
                               max_length=512,
                               truncation=True)
                                     
     output = model.generate(input_ids = encoded_input.input_ids,
                             attention_mask = encoded_input.attention_mask)
     output_answer = tokenizer.decode(output[0], skip_special_tokens=True)

     result_str = "# ANSWER "+str(count)+": "+ output_answer +"\n"  
     result_str = result_str + "REFERENCE: "+ evidence_sentence + "\n\n"
     top_5_query_paragraph_answer_list += result_str
     count+=1
     
    return top_5_query_paragraph_answer_list


query = st.text_area("Query", "", height=25)
document = st.text_area("Document Text", "", height=100)

if st.button("Get Answers From Document"):
  st.markdown(fetch_answers(query, document))