EL-Attention: Memory Efficient Lossless Attention for Generation
Abstract
Transformer model with multi-head attention requires caching intermediate results for efficient inference in generation tasks. However, cache brings new memory-related costs and prevents leveraging larger batch size for faster speed. We propose memory-efficient lossless attention (called EL-attention) to address this issue. It avoids heavy operations for building multi-head keys and values, cache for them is not needed. EL-attention constructs an ensemble of attention results by expanding query while keeping key and value shared. It produces the same result as multi-head attention with less GPU memory and faster inference speed. We conduct extensive experiments on Transformer, BART, and GPT-2 for summarization and question generation tasks. The results show EL-attention speeds up existing models by 1.6x to 5.3x without accuracy loss.
Cite
@article{arxiv.2105.04779,
title = {EL-Attention: Memory Efficient Lossless Attention for Generation},
author = {Yu Yan and Jiusheng Chen and Weizhen Qi and Nikhil Bhendawade and Yeyun Gong and Nan Duan and Ruofei Zhang},
journal= {arXiv preprint arXiv:2105.04779},
year = {2021}
}
Comments
ICML 2021. Version 2: add pseudocode