SiBert Chinese

September 6, 2019 ยท View on GitHub

Introduction

The Sentence Insertion based Bert (SiBert) is pretrained with the the sentence insertion task.

sibert

Update Notes

  • Sep 6, 2019: Initial release for SiBert code.

Codes

  1. create_pretraining_data.py

Create the pretraining data for SiBert.

  1. pretrain.py

Pretrain the SiBert.

  1. run_finetune_cmrc201x.py

Finetune the SiBert for cmrc2018 and cmrc2019 tasks.