Author: Ashish Kumar, Head of Platforms, AI & Data at Tata Group · ashishkumar.ai
Use this model to generate variations to augment the training data used for NLU systems.
from transformers import AutoTokenizer, AutoModelWithLMHead
import torch
if torch.cuda.is_available():
device = torch.device("cuda")
else :
device = "cpu"
tokenizer = AutoTokenizer.from_pretrained("Ashishkr/Gpt2-paraphrase_generation")
model = AutoModelWithLMHead.from_pretrained("Ashishkr/Gpt2-paraphrase_generationn").to(device)
input_query="every moment is a fresh beginning"
query= input_query + " ~~ "
input_ids = tokenizer.encode(query.lower(), return_tensors='pt').to(device)
sample_outputs = model.generate(input_ids,
do_sample=True,
num_beams=1,
max_length=128,
temperature=0.9,
top_p= 0.99,
top_k = 30,
num_return_sequences=40)
paraphrases = []
for i in range(len(sample_outputs)):
r = tokenizer.decode(sample_outputs[i], skip_special_tokens=True).split('||')[0]
r = r.split(' ~~ ')[1]
if r not in paraphrases:
paraphrases.append(r)
print(paraphrases)
To evaluate if a paraphrase is a semantic variation to the input query or just a surface level variation & rank the generated paraphrases, use the following model: