Hugging Face's logo

Spaces:
Vaishakhh
/
Word_Paraphraser_Seq-2-Seq
Running

App Files Community
1
Word_Paraphraser_Seq-2-Seq / app.py
Vaishakhh's picture
Vaishakhh
Update app.py
f677b84 verified
raw history blame contribute delete
No virus
3.21 kB
# -*- coding: utf-8 -*-
"""Ai Re-Phraser.py
Automatically generated by Colaboratory.
Original file is located at
https://colab.research.google.com/drive/18bvmXQqMIkk7G0gY_1dUolI08RK6Ajrf
"""
# importing the libraries
import os
import pandas as pd
from parrot import Parrot #Parrot offers knobs to control Adequacy, Fluency and Diversity metrics
import torch
import warnings
import nltk
import warnings
warnings.filterwarnings("ignore")
from sentence_splitter import SentenceSplitter, split_text_into_sentences
splitter = SentenceSplitter(language='en')
from parrot import Parrot
from transformers import PegasusForConditionalGeneration, PegasusTokenizer
from transformers import AutoTokenizer
from transformers import AutoModelForSeq2SeqLM
from parrot.filters import Adequacy
from parrot.filters import Fluency
from parrot.filters import Diversity
# Adding the metrics
adequacy_score = Adequacy()
fluency_score = Fluency()
diversity_score= Diversity()
device= "cuda:0"
adequacy_threshold = 0.90# Adequacy of the English sentences
fluency_threshold = 0.80 # English Fluency
diversity_ranker="levenshtein" # Diversity (Lexical / Phrasal / Syntactical) (How much has the paraphrase changed the original sentence?)
do_diverse=False # Diverse The sentences formation
#num_beam_groups (int) — Number of groups to divide num_beams into in order to ensure diversity among different groups of beams
# adding the model
model_name = 'tuner007/pegasus_paraphrase'
torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'
tokenizer = PegasusTokenizer.from_pretrained(model_name) #Pre-trained Model
model_pegasus = PegasusForConditionalGeneration.from_pretrained(model_name).to(torch_device)
def get_max_str(lst):
return max(lst, key=len)
def get_response(input_text):
batch = tokenizer.prepare_seq2seq_batch([input_text],truncation=True,padding='longest',max_length=90, return_tensors='pt').to(torch_device)
translated = model_pegasus.generate(**batch,max_length=30,num_beams=15, num_return_sequences=15, temperature=1.5)
#num_beam_groups=num_beams, diversity_penalty=0.5
tgt_text = tokenizer.batch_decode(translated, skip_special_tokens=True)
try:
adequacy_filtered = adequacy_score.filter(input_text,tgt_text, adequacy_threshold, device)
if len(adequacy_filtered) > 0 :
fluency_filtered = fluency_score.filter(adequacy_filtered, fluency_threshold, device )
if len(fluency_filtered) > 0 :
diversity_scored = diversity_score.rank(input_text, fluency_filtered, diversity_ranker)
return get_max_str(diversity_scored)
else:
return get_max_str(fluency_filtered)
else:
return get_max_str(adequacy_filtered)
except:
return(get_max_str(tgt_text))
# Deploying the model
import gradio as gr
def get_fun(Input_txt):
tokens = splitter.split(text=Input_txt)
txt_paraphrase=''
for phrase in tokens:
tmp=get_response(phrase)
txt_paraphrase=txt_paraphrase+' '+tmp
return txt_paraphrase
iface = gr.Interface(fn=get_fun, inputs="text", outputs="text", title = "Ai Re-Phraser")
iface.launch(inline=False)
"""# New Section"""