Quantization Robust Federated Learning for Efficient Inference on Heterogeneous Devices.