NVIDIA's highly optimized transformer-based XLNet and Transformer-XL implementation written in cuda and supports memory mechanism. - View it on GitHub