5fa1a76
1
2
In PyTorch and Tensorflow, this can be done by replacing them with -100, which is the ignore_index of the CrossEntropyLoss.
