DataLoader, when num_worker >0, there is bug

15

so then it seems to be a problem relating to hdf5 in general in combination with multirpocessing rather than with Pytorch itself? If so, it would be interesting so see, how this issue evolves when using the C++ frontend