TinyBERT: Distilling BERT for Natural Language Understanding
Xiaoqi Jiao | Yichun Yin | Lifeng Shang | Xin Jiang | Xiao Chen | Linlin Li | Fang Wang | Qun Liu
Findings of the Association for Computational Linguistics: EMNLP 2020
Language model pre-training, such as BERT, has significantly improved the performances of many natural language processing tasks. However, pre-trained language models are usually computationally expensive, so it is difficult to efficiently execute them on resource-restricted devices. To accelerate inference and reduce model size while maintaining accuracy, we first propose a novel Transformer distillation method that is specially designed for knowledge distillation (KD) of the Transformer-based models. By leveraging this new KD method, the plenty of knowledge encoded in a large “teacher” BERT can be effectively transferred to a small “student” TinyBERT. Then, we introduce a new two-stage learning framework for TinyBERT, which performs Transformer distillation at both the pre-training and task-specific learning stages. This framework ensures that TinyBERT can capture the general-domain as well as the task-specific knowledge in BERT. TinyBERT4 with 4 layers is empirically effective and achieves more than 96.8% the performance of its teacher BERT-Base on GLUE benchmark, while being 7.5x smaller and 9.4x faster on inference. TinyBERT4 is also significantly better than 4-layer state-of-the-art baselines on BERT distillation, with only ~28% parameters and ~31% inference time of them. Moreover, TinyBERT6 with 6 layers performs on-par with its teacher BERT-Base.
This paper proposes a simple CNN model for creating general-purpose sentence embeddings that can transfer easily across domains and can also act as effective initialization for downstream tasks. Recently, averaging the embeddings of words in a sentence has proven to be a surprisingly successful and efficient way of obtaining sentence embeddings. However, these models represent a sentence, only in terms of features of words or uni-grams in it. In contrast, our model (CSE) utilizes both features of words and n-grams to encode sentences, which is actually a generalization of these bag-of-words models. The extensive experiments demonstrate that CSE performs better than average models in transfer learning setting and exceeds the state of the art in supervised learning setting by initializing the parameters with the pre-trained sentence embeddings.