This code contains automatic forward-backward propagation for Long short-term memory (LSTM). Its performance is optimized using CUDA to allow for parallelism. Object-oriented design provides straightforward training process and architecture selection.