SRU (Simple Recurrent Units) is a high-performance recurrent unit library for PyTorch that achieves significant speed-ups over standard LSTM by enabling highly parallelizable computation. The library implements both the original SRU architecture from EMNLP 2018 and the newer SRU++ variant from 2021, which combines attention mechanisms with fast recurrence. It supports installation via pip or source and requires PyTorch and ninja build system.