This is a really simple little encoding with d_vocab = 14 (0-9, +, -, =, >), where '>' is the BOS/EOS token.
Models trained with this should be incoming soon-ish.
