Bug when evaluating the model with a single data point

The OneHotEncoding.forward method encodes the tensor [[3]] as [0, 0, 1], rather than [[0, 0, 1]].