DistributedAdam

class paddle.fluid.incubate.fleet.parameter_server.pslib.optimizer_factory. DistributedAdam ( optimizer ) [source]

adam optimizer in distributed training

minimize ( losses, startup_program=None, parameter_list=None, no_grad_set=None )
Parameters
  • losses (Variable) – loss variable defined by user

  • startup_program (Program) – startup program that defined by user

  • parameter_list (str list) – parameter names defined by users

  • no_grad_set (set) – a set of variables that is defined by users so that these variables do not need gradient computation