mindx_elastic.restore_module.restore_manager.restore_manager.RestoreCheckpoint

restore_exception_checkpoint(CheckpointRestoreParas)

Loads the dying gasp checkpoint of the restoration policy for hybrid parallel training jobs.

**Table 1** CheckpointRestoreParas structure
Item	Type	Description
args_param	NameSpace	Parameter configuration for the training jobs
sink_size	int	Sink size
dataset	mindspore.dataset: MindSpore data type	Dataset used for training
model	mindspore.train.model: MindSpore training type	Train the model.
network	mindspore.nn.cell: MindSpore cell type	Model network used for training
epoch	int	Number of training epochs.

**Table 2** Return value
Parameter	Type	Description
bool	boolean	Indicates whether the checkpoint is successfully loaded after training recovery

Example:

from mindx_elastic.restore_module.restore_manager.restore_checkpoint import RestoreCheckpoint, CheckpointRestoreParas
res_ckpt = RestoreCheckpoint()
input_checkpoint_paras = CheckpointRestoreParas(args_opt, args_opt.sink_size, ds, model,pangu_alpha_with_grads, epoch=actual_epoch_num)
flag = res_ckpt.restore_exception_checkpoint(input_checkpoint_paras)

Parent topic: Elastic-Agent (APIs Related to Resumable Training)