Poplar: Efficient Scaling of Distributed DNN Training on Heterogeneous GPU Clusters

Re-assign community