app.py

from __future__ import absolute_import, division, print_function, unicode_literals
import os
import gradio as gr
from fastai.text.all import *
from transformers import *
from blurr.data.all import *
from blurr.modeling.all import *
import spacy
from spacy_readability import Readability

readablility_nlp = spacy.load('en_core_web_sm')
read = Readability()
cwd = os.getcwd()
readablility_nlp.add_pipe(read, last=True)

bart_ext_model_path = os.path.join(cwd, 'models/bart_extractive_model')
bart_extractive_model = BartForConditionalGeneration.from_pretrained(bart_ext_model_path)
bart_extractive_tokenizer = BartTokenizer.from_pretrained('facebook/bart-large-cnn')

t5_model_path = os.path.join(cwd, 'models/t5_model')
t5_model = AutoModelWithLMHead.from_pretrained(t5_model_path)
t5_tokenizer = AutoTokenizer.from_pretrained("mrm8488/t5-base-finetuned-summarize-news")

def generate_text_summarization(sum_type,article):

    if sum_type == 'BART Extractive Text Summarization':
        inputs = bart_extractive_tokenizer([article], max_length=1024, return_tensors='pt')
        summary_ids = bart_extractive_model.generate(inputs['input_ids'], num_beams=4, min_length=60, max_length=300, early_stopping=True)

        summary = [bart_extractive_tokenizer.decode(g, skip_special_tokens=True, clean_up_tokenization_spaces=False) for g in summary_ids]
        print(type(summary))
        print(summary)
        summary= summary[0] 
        doc = readablility_nlp(summary)
        summary_score = round(doc._.flesch_kincaid_reading_ease,2)
        summarized_data = {
            "summary" : summary,
            "score" : summary_score
        }
        return summary

    if sum_type == 'T5 Abstractive Text Summarization':
        inputs = t5_tokenizer.encode(article, return_tensors="pt", max_length=2048)
        summary_ids = t5_model.generate(inputs,
                                num_beams=2,
                                no_repeat_ngram_size=2,
                                min_length=100,
                                max_length=300,
                                early_stopping=True)

        summary = t5_tokenizer.decode(summary_ids[0], skip_special_tokens=True)
        print(type(summary))
        print(summary)
        doc = readablility_nlp(summary)
        summary_score = round(doc._.flesch_kincaid_reading_ease,2)
        summarized_data = {
            "summary" : summary,
            "score" : summary_score
        }
        return summary

input_text=gr.Textbox(lines=5, label="Paragraph")
input_radio= gr.Radio(['BART Extractive Text Summarization','T5 Abstractive Text Summarization'],label='Select summarization',value='BART Extractive Text Summarization')
output_text=gr.Textbox(lines=7, label="Summarize text")
demo = gr.Interface(
    generate_text_summarization,
    [input_radio,input_text],
    output_text,
    title="Text Summarization",
    css=".gradio-container {background-color: lightgray}",
    article="""<p style='text-align: center;'>Developed by: <a href="https://www.pragnakalp.com" target="_blank">Pragnakalp Techlabs</a></p>"""
)

demo.launch()