Public API for tf._api.v2.train.experimental namespace
Classes
class DynamicLossScale
: Loss scale that dynamically adjusts itself.
class FixedLossScale
: Loss scale with a fixed value.
class LossScale
: Base class for all TF1 loss scales.
class MaxShardSizePolicy
: Policy that splits tensors into shards with a max shard size.
class MixedPrecisionLossScaleOptimizer
: An optimizer that applies loss scaling.
class PythonState
: A mixin for putting Python state in an object-based checkpoint.
class ShardByTaskPolicy
: Policy that splits tensors into shards based on their device spec task.
class ShardableTensor
: Tensor wrapper containing data necessary for sharding.
class ShardingCallback
: Checkpoint sharding callback function, along with a text description.
Functions
disable_mixed_precision_graph_rewrite(...)
: Disables the mixed precision graph rewrite.
enable_mixed_precision_graph_rewrite(...)
: Enable mixed precision via a graph rewrite.