Timer_Trans_DE.py

import pandas as pd
import torch
import numpy as np
from transformers import pipeline
import time
from transformers import MarianTokenizer, MarianMTModel
from typing import List

print("Enter attack csv file name")
path=input()

ds1=pd.read_csv(path)


## Translation pipeline
src = 'en'  # source language
trg = 'de'  # target language
mname = f'Helsinki-NLP/opus-mt-{src}-{trg}'
model = MarianMTModel.from_pretrained(mname)
tok = MarianTokenizer.from_pretrained(mname)
def en_de(txt):
  batch = tok.prepare_seq2seq_batch(src_texts=[txt])  # don't need tgt_text for inference
  gen = model.generate(**batch)  # for forward pass: model(**batch)
  words: List[str] = tok.batch_decode(gen, skip_special_tokens=True)  # returns "Where is the bus stop ?"
  return(words[0])

model_name = f'Helsinki-NLP/opus-mt-{trg}-{src}'
tokenizer = MarianTokenizer.from_pretrained(model_name)
model1 = MarianMTModel.from_pretrained(model_name)
def de_en(txtx):
  translated = model1.generate(**tokenizer.prepare_seq2seq_batch([txtx]))
  tgt = [tokenizer.decode(t, skip_special_tokens=True) for t in translated]
  return(tgt)

from transformers import AutoTokenizer, AutoModelForSequenceClassification

tok_name = input("Enter Tokenizer name (corr. to the classifier model) from Hugging Face")
tokenizer = AutoTokenizer.from_pretrained(tok_name)

model_name = input("Enter classifier model name from Hugging Face")
model = AutoModelForSequenceClassification.from_pretrained("model_name")

time_tran=[]
time_cls=[]


list1=ds1.perturbed_text
list2=[]
a=0

for x in list1[:20]:
  cnta=time.time()
  w1 = en_de(x)
  list2.append(de_en(w1))
  cntb=time.time()
  time_tran.append(cntb-cnta)
  print(f"Example number {a}")
  a+=1


translated=list2
perturbed=list1
initial_prob=list(ds1.original_output)
initial_score=list(ds1.original_score)
perturbed_prob=list(ds1.perturbed_output)
perturbed_score=list(ds1.perturbed_score)

final_prob=[]
final_score=[]



a=0
for y in translated:
  cnta= time.timeit()
  txf1=tokenizer.encode_plus(y,return_tensors="pt",max_length=512)
  txf2=model(**txf1)[0]
  results=torch.softmax(txf2, dim=1).tolist()[0]
  final_prob.append(np.argmax(results))
  final_score.append(results[np.argmax(results)])
  cntb=time.timeit()
  time_tran[a]=time_sum[a]+(cntb-cnta)
  print(f"Iteration number={a}")
  a+=1

a=0
for y in perturbed[:20]:
  cnta= time.timeit()
  txf1=tokenizer.encode_plus(y,return_tensors="pt",max_length=512)
  txf2=model(**txf1)[0]
  results=torch.softmax(txf2, dim=1).tolist()[0]
  final_prob.append(np.argmax(results))
  final_score.append(results[np.argmax(results)])
  cntb=time.timeit()
  time_cls.append(cntb-cnta)
  print(f"Iteration number={a}")
  a+=1


print(f"Average time w/o defense={np.mean(time_cls)}")
print(f"Average time w defense_German={np.mean(time_tran)}")