In natural language processing, pre-trained language models have become essential infrastructures. However, these models often suffer from issues such as large size, long inference time, and challenging deployment. Moreover, most mainstream pre-trained models focus on English, and there are insufficient studies on small Chinese pre-trained models. In this paper, we introduce MiniRBT, a small Chinese pre-trained model that aims to advance research in Chinese natural language processing. MiniRBT employs a narrow and deep student model and incorporates whole word masking and two-stage distillation during pre-training to make it well-suited for most downstream tasks. Our experiments on machine reading comprehension and text classification tasks reveal that MiniRBT achieves 94% performance relative to RoBERTa, while providing a 6.8x speedup, demonstrating its effectiveness and efficiency.
@article{arxiv.2304.00717,
title = {MiniRBT: A Two-stage Distilled Small Chinese Pre-trained Model},
author = {Xin Yao and Ziqing Yang and Yiming Cui and Shijin Wang},
journal= {arXiv preprint arXiv:2304.00717},
year = {2023}
}