In NLP char generation, if we use one$-$hot encoding for each char, which loss function should we use?

Group of answer choices

softmax

Sparse categorical cross entropy

binary cross entropy

Mean square error