ModelCheckpoint

classin packageklyn.math.nn.serialization

public class ModelCheckpoint<T extends AbstractFloat>

ModelCheckpoint

Private transport owned by the checkpoint loader until all bytes validate. It never references objects in the model being replaced.

Example

checkpoint = ModelCheckpoint<Float>()

Fields

Modifier and Type	Member	Description
public	finiteBatches `finiteBatches as Long = 0l`	Finite-batch count for scale growth.
public	gradientScale `gradientScale as Double = 1.0`	FP16 dynamic loss scale.
public	inputShape `inputShape as IList<Long> = []`	Checked input sample shape.
public	layers `layers as ArrayList<Layer<T>> = []`	Restored stages.
public	loss `loss as ClassificationLoss<T> = null`	Classification loss, or null for an uncompiled model.
public	optimizer `optimizer as Optimizer<T> = null`	Optimizer or null for an inference-only checkpoint.
public	regressionLoss `regressionLoss as RegressionLoss<T> = null`	Regression objective, mutually exclusive with loss.