Assi Jakoby

The above gradient accumulation implementation doesn;t work with TF2.5 with multi GPU distribution strategy strategy = tf.distribute.MultiWorkerMirroredStrategy()

> ```python > import tensorflow as tf > from tensorflow_addons.utils import types > from typeguard import typechecked > > class GradientAccumulator(tf.keras.optimizers.Optimizer): > """Gradient accumulation utility. > When used with a...