INNER CODE UNIT · Python

PartialDistributedOptimizer

horovod/horovod · horovod/keras/__init__.py:116

def PartialDistributedOptimizer(optimizer, name=None,
                                device_dense='', device_sparse='',
                                compression=Compression.none,
                                sparse_as_dense=False,
                                gradient_predivide_factor=1.0,
                                op=Average,
                                backward_passes_per_step=1,
                                average_aggregated_gradients=False,
                                groups=None,
                                process_set=global_process_set,
                                local_layers=None, scale_local_gradients=True):
    """
    An optimizer that wraps another keras.optimizers.Optimizer, using an allreduce to
    average gradient values before applying gradients to model weights.

    Args:
        optimizer: Optimizer to use for computing gradients and applying updates.
        process_set: Gradients will only be reduced over Horovod processes belonging

View source record →