app.py

import pickle
import gradio as gr
import torch
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
from transformers import BertTokenizer, BertForSequenceClassification, pipeline, AutoTokenizer, AutoModelForSequenceClassification, TextClassificationPipeline, AutoModelForSeq2SeqLM, AutoModel, RobertaModel, RobertaTokenizer
from sentence_transformers import SentenceTransformer
from fin_readability_sustainability import BERTClass, do_predict
import pandas as pd

#import lightgbm
#lr_clf_finbert = pickle.load(open("lr_clf_finread_new.pkl",'rb'))
tokenizer_read = BertTokenizer.from_pretrained('ProsusAI/finbert')


device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model_read = BERTClass(2, "readability")
model_read.to(device)
model_read.load_state_dict(torch.load('readability_model.bin', map_location=device)['model_state_dict'], strict=False)


def get_readability(text):
  df = pd.DataFrame({'sentence':[text]})
  actual_predictions_read = do_predict(model_read, tokenizer_read, df)
  score = round(actual_predictions_read[1][0], 4)
  return score

# Reference : https://huggingface.co/humarin/chatgpt_paraphraser_on_T5_base
tokenizer = AutoTokenizer.from_pretrained("humarin/chatgpt_paraphraser_on_T5_base")
model = AutoModelForSeq2SeqLM.from_pretrained("humarin/chatgpt_paraphraser_on_T5_base")

def paraphrase(
    question,
    num_beams=5,
    num_beam_groups=5,
    num_return_sequences=5,
    repetition_penalty=10.0,
    diversity_penalty=3.0,
    no_repeat_ngram_size=2,
    temperature=0.7,
    max_length=128
):
    input_ids = tokenizer(
        f'paraphrase: {question}',
        return_tensors="pt", padding="longest",
        max_length=max_length,
        truncation=True,
    ).input_ids
    
    outputs = model.generate(
        input_ids, temperature=temperature, repetition_penalty=repetition_penalty,
        num_return_sequences=num_return_sequences, no_repeat_ngram_size=no_repeat_ngram_size,
        num_beams=num_beams, num_beam_groups=num_beam_groups,
        max_length=max_length, diversity_penalty=diversity_penalty
    )

    res = tokenizer.batch_decode(outputs, skip_special_tokens=True)

    return res

def get_most_readable_paraphrse(text):
  li_paraphrases = paraphrase(text)
  li_paraphrases.append(text)
  best = li_paraphrases[0]
  score_max = get_readability(best)
  for i in range(1,len(li_paraphrases)):
    curr = li_paraphrases[i]
    score = get_readability(curr)
    if score > score_max:
      best = curr
      score_max = score
  if best!=text and score_max>.6:
    ans = "The most redable version of text that I can think of is:\n" + best  
  else:
    "Sorry! I am not confident. As per my best knowledge, you already have the most readable version of the text!"
  return ans

def set_example_text(example_text):
    return gr.Textbox.update(value=example_text[0])

with gr.Blocks() as demo:
    gr.Markdown(
    """
    # FinLanSer
    Financial Language Simplifier
    """)
    text = gr.Textbox(label="Enter text you want to simply (make more readable)")
    greet_btn = gr.Button("Simplify/Make Readable")
    output = gr.Textbox(label="Output Box")
    greet_btn.click(fn=get_most_readable_paraphrse, inputs=text, outputs=output, api_name="get_most_raedable_paraphrse")
    example_text = gr.Dataset(components=[text], samples=[['Legally assured line of credit with a bank'], ['A mutual fund is a type of financial vehicle made up of a pool of money collected from many investors to invest in securities like stocks, bonds, money market instruments']])
    example_text.click(fn=set_example_text, inputs=example_text,outputs=example_text.components)  

demo.launch()