paraphrase.py

from transformers import PegasusForConditionalGeneration, PegasusTokenizer

class PegasusParaphraser:
    """ Pegasus Model for Paraphrase"""

    def __init__(self, num_return_sequences=3, num_beams=10, max_length=60,temperature=1.5, device="cpu"):
        self.model_name = "tuner007/pegasus_paraphrase"
        self.device = device
        self.model = self.load_model()
        self.tokenizer = PegasusTokenizer.from_pretrained(self.model_name)
        self.num_return_sequences = num_return_sequences
        self.num_beams = num_beams
        self.max_length=max_length
        self.temperature=temperature


    def load_model(self):
        model = PegasusForConditionalGeneration.from_pretrained(self.model_name).to(self.device)
        return model


    def paraphrase(self,input_text ):

        batch = self.tokenizer(
            [input_text],
            truncation=True,
            padding="longest",
            max_length=self.max_length,
            return_tensors="pt",
        ).to(self.device)
        translated = self.model.generate(
            **batch,
            max_length=self.max_length,
            num_beams=self.num_beams,
            num_return_sequences=self.num_return_sequences,
            temperature=self.temperature
        )
        tgt_text = self.tokenizer.batch_decode(translated, skip_special_tokens=True)
        return tgt_text