tamakoji / pytorch-syncbn

Synchronized Multi-GPU Batch Normalization

tamakoji/pytorch-syncbn Issues

Error about BatchNorm2dSync for 1d input in eval model, but I can use this BatchNorm2dSync for 1d input in train model
Closed 3 years ago
I have not seen any differences between m1 and m2, Why?
Updated 4 years ago2
subprocess.CalledProcessError: Command '['which', 'c++']' returned non-zero exit status 1
Updated 4 years ago2
support for fp16
Updated 4 years ago
Please make this into a pip package.
Updated 5 years ago
Does this support 3dBN?
Updated 5 years ago
I get stuck after several iterations
Closed 5 years ago
The gpu memory increase when number of gpus > 4.
Updated 5 years ago
Gpu memory problem. In training phase, it will gradually occupy all the gpu memory.
Closed 5 years ago13
RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation
Closed 5 years ago
Torch 1.0.1 compatibility
Closed 5 years ago1
Can it use in 3d？
Closed 6 years ago