Translations:PyTorch/306/en
Jump to navigation
Jump to search
There is a known issue with our PyTorch 1.10 wheel torch-1.10.0+computecanada. Multi-GPU code that uses DistributedDataParallel running with this PyTorch version may fail unpredictably if the backend is set to 'nccl' or 'gloo'. We recommend using our latest PyTorch build instead of version 1.10 on all GP clusters.