CAB: Comprehensive Attention Benchmarking on Long Sequence Modeling
Published in ICLR, 2022
CAB aims to provide a comprehensive evaluation of efficient attentions. CAB contains seven real-world tasks from different research areas to evaluate efficient attentions under four fine-grained attention patterns.
Recommended citation: @article{zhang2022cab, title={CAB: Comprehensive Attention Benchmarking on Long Sequence Modeling}, author={Zhang, Jun and Jiang, Shuyang and Feng, Jiangtao and Zheng, Lin and Kong, Lingpeng}, journal={arXiv preprint arXiv:2210.07661}, year={2022} } https://arxiv.org/pdf/2210.07661.pdf