Multiprocessing DataLoader is getting killed #5
Description
If num_workers in DataLoader is more than 0 (it uses multiprocessing). The training process can sometimes get randomly killed. Possible issue: some of the operations in the collate or get item function aren't parallelable?