DistributedReplicatedParameters

class deepinv.distributed.framework.DistributedReplicatedParameters(ctx, parameters, average=True)

Bases: object

Synchronize gradients for replicated trainable parameters.

This class targets parameters that are replicated on all ranks (e.g. trainable step sizes in unrolled algorithms) and are not otherwise handled by DistributedProcessing.