sharad/ParaphraseGPT

This model re-fine-tunes the ChatGPT Paraphraser on T5 Base with additional Google PAWS dataset.

Usage example

from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLM

#'cuda' for gpu otherwise use 'cpu'
device = "cuda"
model     = AutoModelForSeq2SeqLM.from_pretrained("sharad/ParaphraseGPT").to(device)
tokenizer = AutoTokenizer.from_pretrained("humarin/chatgpt_paraphraser_on_T5_base")
predict   = pipeline("text2text-generation", model=model, tokenizer=tokenizer)

def paraphrase(sentence):
  generated = predict(
              sentence,
              num_beams=3,
              num_beam_groups=3,
              num_return_sequences=1,
              diversity_penalty=2.0,
              no_repeat_ngram_size=2,
              repetition_penalty=0.99,
              max_length=len(sentence)
          )
  return generated

output = paraphrase('My sentence to paraphrase...')
print(output[0]['generated_text'])

Train parameters

epochs = 4
max_length = 128
lr = 5e-5