As if we didn’t need more reasons to get rid of the GIL, this deep-dive analysis demonstrates how to use process-based async to work around the GIL when training models with PyTorch. No word about use ...