The compression ratio is determined by two predefined hyperparameters: the method of compression (channel/block-level pruning or quantization) and latency loss alpha.
As of now, we should set these hyperparameters manually depending on our experience.
To produce many different tradeoffs automatically, an automatic hyperparameter selector should be implemented.
The compression ratio is determined by two predefined hyperparameters: the method of compression (channel/block-level pruning or quantization) and latency loss alpha.
As of now, we should set these hyperparameters manually depending on our experience.
To produce many different tradeoffs automatically, an automatic hyperparameter selector should be implemented.