Fetching the paper…

TernGrad: Ternary Gradients to Reduce Communication in Distributed Deep Learning · Around