This repo aims at providing a collection of efficient Triton-based implementations for state-of-the-art linear attention models. All implementations are written purely in PyTorch and Triton, making ...

All the benefits of plugins with none of the downsides.