Recurrent Neural Networks (RNNs) are powerful tools for solving sequence-based problems, but their efficacy and execution time are dependent on the size of the network. Following recent work in simplifying these networks with model pruning and a novel mapping of work onto GPUs, we design an efficient implementation for sparse RNNs. We investigate several optimizations and tradeoffs: Lamport timestamps, wide memory loads, and a bank-aware weight layout. With these optimizations, we achieve speedups of over 6x over the next best algorithm for a hidden layer of size 2304, batch size of 4, and a density of 30%. Further, our technique allows for models of over 5x the size to fit on a GPU for a speedup of 2x, enabling larger networks to help advance the state-of-the-art. We perform case studies on NMT and speech recognition tasks in the appendix, accelerating their recurrent layers by up to 3x.
@article{arxiv.1804.10223,
title = {Sparse Persistent RNNs: Squeezing Large Recurrent Networks On-Chip},
author = {Feiwen Zhu and Jeff Pool and Michael Andersch and Jeremy Appleyard and Fung Xie},
journal= {arXiv preprint arXiv:1804.10223},
year = {2018}
}