A simple implementation of LSTM from scratch in Python, then dump trained parameters to a .coe file, the goal is load data to FPGA RAMs. The parameters can be loaded and used to predict the next character in a sequence.
The code is based on the blog post by Christopher Olah.