INNER CODE UNIT · Python

_DistributedOptimizer

bytedance/byteps · byteps/_keras/__init__.py:22

    class _DistributedOptimizer(keras.optimizers.Optimizer):
        _HAS_AGGREGATE_GRAD = True
        def __init__(self, **kwargs):
            self._name = name or "Distributed%s" % self.__class__.__base__.__name__
            self._device_dense = device_dense
            self._device_sparse = device_sparse
            self._compression = compression
            self._sparse_as_dense = sparse_as_dense
            self._aggregated_gradients = False
            super(self.__class__, self).__init__(**kwargs)

        def get_gradients(self, loss, params):
            """
            Compute gradients of all trainable variables.
            See Optimizer.get_gradients() for more info.
            In DistributedOptimizer, get_gradients() is overriden to also
            push_pull the gradients before returning them.
            """

View source record →