Loss always equal to zero while training the model

Weighted sampling for multi-labels classification is not trivial, since each sample would contain zero, one, or multiple classes. This post mentions a few approaches which you could check.

1 Like