PyTorch

For multiple GPUs, use distributed data parallel

Related libraries