Collected sources and patterns will appear here. Add from search, explore, or the patterns library.
GradientTensor -> CompressedSketch
Compress client model gradient updates using a Count-Sketch data structure before transmission to reduce communication bandwidth.
Problem it solves
High communication overhead when transmitting raw neural network parameters in distributed training.
Consumes
Emits
The real projects this mechanism was found in. Attribution is the point — this is how the best teams actually do it.