ERNIE-tiny

Introduction

ERNIE-tiny is a compressed model from ERNIE 2.0 base model through model structure compression and model distillation. Through compression, the performance of the ERNIE-tiny only decreases by an average of 2.37% compared to ERNIE 2.0 base, but it outperforms Google BERT by 8.35%, and the speed increases by 4.3 times.

More details: https://github.com/PaddlePaddle/ERNIE/blob/develop/distill/README.md

Released Model Info

This released pytorch model is converted from the officially released PaddlePaddle ERNIE model and a series of experiments have been conducted to check the accuracy of the conversion.

• Official PaddlePaddle ERNIE repo: https://github.com/PaddlePaddle/ERNIE
• Pytorch Conversion repo: https://github.com/nghuyong/ERNIE-Pytorch

How to use

from transformers import AutoTokenizer, AutoModel
tokenizer = AutoTokenizer.from_pretrained("nghuyong/ernie-tiny")
model = AutoModel.from_pretrained("nghuyong/ernie-tiny")

Citation

@article{sun2019ernie20,
  title={ERNIE 2.0: A Continual Pre-training Framework for Language Understanding},
  author={Sun, Yu and Wang, Shuohuan and Li, Yukun and Feng, Shikun and Tian, Hao and Wu, Hua and Wang, Haifeng},
  journal={arXiv preprint arXiv:1907.12412},
  year={2019} 
}