TypeError when modifying tf.keras.Optimizer parameter

Question

I have been trying to implement a (heavily-inspired) Gradient-Accumulation wrapper for my Adam optimizer, as shown below:

class AccumOptimizer(tf.keras.optimizers.Optimizer):

    def __init__(self, optimizer, steps_per_update=1, **kwargs):

        super(AccumOptimizer, self).__init__(name="AccumOptimizer", **kwargs)
        self.optimizer = optimizer
        self.steps_per_update = steps_per_update
        self.iterations = tf.Variable(0, dtype='int64', name='iterations')
        self.cond = tf.equal(self.iterations % self.steps_per_update, 0)
        self.lr = self.optimizer.learning_rate
        self.optimizer.learning_rate = tf.cond(self.cond,
                                               lambda: self.optimizer.learning_rate,
                                               lambda: tf.constant(0, tf.float32))
        ...

But seem to butt into the following TypeError when I try to train:

TypeError: __array__() takes 1 positional argument but 2 were given

which points to the part where self.optimizer.learning_rate is updated based on self.cond, although I'm sure that tf.condshould return a single value if the true/false function returns a singleton list.

I'm using Tensorflow 1.15.x (and constrained to this version, sadly).

Any ideas how this could be circumvented?