Using one hot encoded labelling, which label is assigned to which number? Does it depend on how you define the variable ‘classes’?
The meaning of the class labels depend on your definition.
I.e. you can create the mapping manually or use any predefined mapping for standard datasets.
However, note that for a vanilla multi-class classification use case, you won’t use one-hot encoded labels, but instead the class indices directly as explained in the nn.CrossEntropyLoss docs.