Text Summarization

In this release, we support both abstractive and extractive text summarization.

New Model: UniLM

UniLM is a state of the art model developed by Microsoft Research Asia (MSRA). The model is pre-trained on a large unlabeled natural language corpus (English Wikipedia and BookCorpus) and can be fine-tuned on different types of labeled data for various NLP tasks like text classification and abstractive summarization.

Supported Models

• unilm-large-cased
• unilm-base-cased

For more info about UniLM, please refer to the following:

• Paper: Unified Language Model Pre-training for Natural Language Understanding and Generation
• Github: https://github.com/microsoft/unilm

Thanks to the UniLM team, Li Dong, Nan Yang, Wenhui Wang, Furu Wei, Xiaodong Liu, Yu Wang, Jianfeng Gao, Ming Zhou, Hsiao-Wuen Hon, for their great work and support for the integration.

New Model: BERTSum

BERTSum is an encoder architecture designed for text summarization. It can be used together with different decoders to support both extractive and abstractive summarization.

Supported Models

• bert-base-uncased (extractive and abstractive)
• distilbert-base-uncased (extractive)

• Papers:

  • Text Summarization with Pretrained Encoders
  • Fine-tune BERT for Extractive Summarization

• GitHub:

  • https://github.com/nlpyang/PreSumm/
  • https://github.com/nlpyang/Bertsum

Thanks to the original authors Yang Liu and Mirella Lapata for their great contribution.

All model implementations support distributed training and multi-GPU inferencing. For abstractive summarization, we also support mixed-precision training and inference.