Builds a replicated input pipeline. This is similar to GenericInput, except that the input processing can be distributed across devices and then concatenated at the current device. Args: processor: see comments for GenericInput. num_replicas: the number of input processing replicas.
(processor, num_replicas, replica_device_fn,
**kwargs)
source not stored for this graph (policy: none)
nothing calls this directly
no test coverage detected
searching dependent graphs…