Pegasus Text Paraphrasing Documentation

The Pegasus model is a powerful tool for text paraphrasing, capable of generating multiple paraphrases for a given input sentence. This documentation provides a guide on how to use the Pegasus model for text paraphrasing and lists the required dependencies.

Table of Contents

1. Introduction
2. Requirements
3. Installation
4. Usage
5. Conclusion

1. Introduction

Pegasus is a state-of-the-art pre-trained model designed for paraphrasing tasks. It leverages the power of deep learning and attention mechanisms to produce high-quality paraphrases for various input sentences. By generating multiple paraphrases, Pegasus provides users with diverse options for text transformation.

2. Requirements

To use the Pegasus model for text paraphrasing, the following dependencies are required:

• absl-py
• mock
• numpy
• rouge-score
• sacrebleu
• sentencepiece
• tensorflow-text==1.15.0rc0
• tensor2tensor==1.15.0
• tensorflow-datasets==2.1.0
• tensorflow-gpu==1.15.2
• sentence-splitter (Install using pip: pip install sentence-splitter)
• transformers (Install using pip: pip install transformers)
• SentencePiece (Install using pip: pip install SentencePiece)

3. Installation

To get started, first install all the required dependencies listed in the previous section.

4. Usage

To use the Pegasus model for paraphrasing, you can use the provided Python code. The code makes use of the Pegasus model and tokenizer from the transformers library.

import torch
from transformers import PegasusForConditionalGeneration, PegasusTokenizer

# Load the pre-trained model and tokenizer
model_name = 'tuner007/pegasus_paraphrase'
torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'

tokenizer = PegasusTokenizer.from_pretrained(model_name)
model = PegasusForConditionalGeneration.from_pretrained(model_name).to(torch_device)

def get_response(input_text, num_return_sequences):
    batch = tokenizer.prepare_seq2seq_batch([input_text],
                                            truncation=True,
                                            padding='longest',
                                            max_length=60, return_tensors="pt").to(torch_device)

    translated = model.generate(**batch, max_length=60,
                                num_beams=10, 
                                num_return_sequences=num_return_sequences, 
                                temperature=1.5)

    tgt_text = tokenizer.batch_decode(translated, skip_special_tokens=True)

    return tgt_text

5. Conclusion

With the Pegasus model, you can easily perform text paraphrasing and explore various alternatives for a given sentence. Enjoy the flexibility and versatility of Pegasus in transforming your text data