apply_weight_standardization#

composer.functional.apply_weight_standardization(module, n_last_layers_ignore=0)[source]#

Weight Standardization standardizes convolutional weights in a model.

Parameters

  • module (Module) โ€“ the torch module whose convolutional weights will be parametrized.

  • n_last_layers_ignore (int, optional) โ€“ the number of layers at the end of the module to not apply weight standardization. Default: 0.