1 paper · 1 filter
Jihao Xin, Ivan Ilin, Shunkang Zhang +2
In distributed training, communication often emerges as a bottleneck. In response, we introduce Kimad, a solution that offers adaptive gradient compression. By consistently monitor…