Learning in Compact Spaces with Approximately Normalized Transformer | Read Paper on Bytez