an issue about CPC code

[zzbb9944]

Hello, thank you for your open-source work on CPC code. But when reading your CPC code, I noticed an issue. When calculating Logsoftmax for zWc, you calculated it by line, that is, for a sample's z_t_k, you thought its c_t was positive, and negative for other samples. However, according to the paper, Logsoftmax should be calculated by column, which means that for a sample's c_t, its z_t_k was positive, and negative for other samples.