Supports 0. regularization. Without this user needs to add an 'if else' when...
Supports 0. regularization. Without this user needs to add an 'if else' when they do hyper parameter tuning for regularization. Currently it crashes if you provide `activity_regularizer=l2(0.)`. Core reason is that L1L2(..)() returns non-tensor object in case of `l=0.`. PiperOrigin-RevId: 234689073
Loading
Please sign in to comment