1 = coded along the input dim (stored [out, in·k/32]); 0 = along the output dim (stored [in, out·k/32], i.e. Wᵀ coded along its last axis).
1 = coded along the input dim (stored [out, in·k/32]); 0 = along the output dim (stored [in, out·k/32], i.e. Wᵀ coded along its last axis).