5fa1a76
1
2
However, it could also be that PyTorch's implementation of a layer requires the weight to be transposed beforehand.
