The codes implement LSTM, GRU RNN models with two rounding methods, including Binarization and Ternarization on FPGA.