DeepLearningExamples
DeepLearningExamples copied to clipboard
[ConvNet/PyTorch] SyncBatchNorm not Used in ConvNet ImageNet Classification
Related to ConvNet/PyTorch (e.g. GNMT/PyTorch or FasterTransformer/All)
Describe the bug SyncBatchNorm is not used in the script https://github.com/NVIDIA/DeepLearningExamples/tree/master/PyTorch/Classification/ConvNets. Is this a potential bug when DDP is activated? I make this guess because I did not see any relevant code in this repo that resolves the out-sync problem between different GPUs.