An error is reported when using keras2.0.8 custom layer, and keras2.3.1 can run normally

Question

Below is my code:

class AttentionOnDims(Layer):
    def __init__(self, **kwargs):
        self.dim = None
        self.kernel = None
        self.built = True