mindspore_lite.ModelGroupFlag
- class mindspore_lite.ModelGroupFlag[source]
The ModelGroupFlag class defines the type of the model group.
The ModelGroupFlag is used to define the flags used to construct a ModelGroup. Currently, supports:
1. ModelGroupFlag.SHARE_WEIGHT, multiple models share weights share workspace memory, default construction flag for ModelGroup.
2. ModelGroupFlag.SHARE_WORKSPACE, multiple models share weights(including constatns and variables) memory. Currently only supported in cloud side Ascend inference and the provider is GE.
Examples
>>> import mindspore_lite as mslite >>> context = mslite.Context() >>> context.target = ["Ascend"] >>> context.ascend.device_id = 0 >>> context.ascend.rank_id = 0 >>> context.ascend.provider = "ge" >>> model_group = mslite.ModelGroup(mslite.ModelGroupFlag.SHARE_WEIGHT) >>> model0 = mslite.Model() >>> model1 = mslite.Model() >>> model_group.add_model([model0, model1]) >>> model0.build_from_file("seq_1024.mindir", mslite.ModelType.MINDIR, context, "config0.ini") >>> model1.build_from_file("seq_1.mindir", mslite.ModelType.MINDIR, context, "config.ini")