Distilling state-of-the-art transformer models into lightweight student models is an effective way to reduce computation cost at inference time. The student models are typically compact transformers with fewer parameters, while expensive operations such as self-attention persist. Therefore, the improved inference speed may still be unsatisfactory for real-time or high-volume use cases. In this paper, we aim to further push the limit of inference speed by distilling teacher models into bigger, sparser student models -- bigger in that they scale up to billions of parameters; sparser in that most of the model parameters are n-gram embeddings. Our experiments on six single-sentence text classification tasks show that these student models retain 97% of the RoBERTa-Large teacher performance on average, and meanwhile achieve up to 600x speed-up on both GPUs and CPUs at inference time. Further investigation reveals that our pipeline is also helpful for sentence-pair classification tasks, and in domain generalization settings.
@article{arxiv.2110.08536,
title = {Sparse Distillation: Speeding Up Text Classification by Using Bigger Student Models},
author = {Qinyuan Ye and Madian Khabsa and Mike Lewis and Sinong Wang and Xiang Ren and Aaron Jaech},
journal= {arXiv preprint arXiv:2110.08536},
year = {2022}
}
Comments
NAACL 2022 camera-ready version. Code: https://github.com/ink-usc/sparse-distillation. In v2, we updated the performance of KD-BiLSTM baselines after fixing a bug