oa::CheckpointManager

CheckpointManager class in the OA Ml public surface.

ClC++ public headers

Public Methods

oa::F64 oa::CheckpointManager::bestMetric()

oa::String oa::CheckpointManager::incrementalDir()

bool oa::CheckpointManager::isBetter(oa::F64 inMetric)

oa::Status oa::CheckpointManager::loadBestInto(oa::Module & inOutModel, oa::Optimizer & inOutOpt)

oa::Status oa::CheckpointManager::loadLatestInto(oa::Module & inOutModel, oa::Optimizer & inOutOpt)

oa::String oa::CheckpointManager::masterPath()

oa::Status oa::CheckpointManager::maybeSave(oa::Module & inModel, oa::Optimizer & inOpt, oa::U64 inStep, oa::F64 inMetric, bool inForce = false)

const oa::String & oa::CheckpointManager::metricName()

oa::String oa::CheckpointManager::modelDir()

oa::Status oa::CheckpointManager::saveIncremental(oa::Module & inModel, oa::Optimizer & inOpt, oa::U64 inStep, oa::F64 inMetric, const oa::String & inMetricName = = {})

Constructor & Destructor Documentation

oa::CheckpointManager::CheckpointManager(
	oa::Engine & inEngine,
	CheckpointManagerConfig inConfig = {}
)

No public source comment is attached to this declaration.

Parameters

inEngine

oa::Engine &

—

inConfig

CheckpointManagerConfig

Default: {}

Public Method Documentation

oa::F64 oa::CheckpointManager::bestMetric()

No public source comment is attached to this declaration.

Returns

oa::F64

The declared return value.

oa::String oa::CheckpointManager::incrementalDir()

No public source comment is attached to this declaration.

Returns

oa::String

The declared return value.

bool oa::CheckpointManager::isBetter(
	oa::F64 inMetric
)

No public source comment is attached to this declaration.

Parameters

inMetric

oa::F64

—

Returns

bool

The declared return value.

oa::Status oa::CheckpointManager::loadBestInto(
	oa::Module & inOutModel,
	oa::Optimizer & inOutOpt
)

Restore weights and optimizer state into an already-constructed model/optimizer. Symmetric with maybeSave.

Parameters

inOutModel

oa::Module &

—

inOutOpt

oa::Optimizer &

—

Returns

oa::Status

The declared return value.

oa::Status oa::CheckpointManager::loadLatestInto(
	oa::Module & inOutModel,
	oa::Optimizer & inOutOpt
)

No public source comment is attached to this declaration.

Parameters

inOutModel

oa::Module &

—

inOutOpt

oa::Optimizer &

—

Returns

oa::Status

The declared return value.

oa::String oa::CheckpointManager::masterPath()

No public source comment is attached to this declaration.

Returns

oa::String

The declared return value.

oa::Status oa::CheckpointManager::maybeSave(
	oa::Module & inModel,
	oa::Optimizer & inOpt,
	oa::U64 inStep,
	oa::F64 inMetric,
	bool inForce = false
)

save model + optimizer state if the metric improved (or unconditionally if inForce=true). Saves weights AND optimizer state (AdamW M/V/step, etc.) into one .oam via Module::save(engine, path, opt).

Parameters

inModel

oa::Module &

—

inOpt

oa::Optimizer &

—

inStep

oa::U64

—

inMetric

oa::F64

—

inForce

bool

Default: false

Returns

oa::Status

The declared return value.

const oa::String & oa::CheckpointManager::metricName()

No public source comment is attached to this declaration.

Returns

const oa::String &

The declared return value.

oa::String oa::CheckpointManager::modelDir()

No public source comment is attached to this declaration.

Returns

oa::String

The declared return value.

oa::Status oa::CheckpointManager::saveIncremental(
	oa::Module & inModel,
	oa::Optimizer & inOpt,
	oa::U64 inStep,
	oa::F64 inMetric,
	const oa::String & inMetricName = = {}
)

save a resumable incremental checkpoint without changing the best-model metric/master file. Used by mid-epoch saves: only a complete epoch metric (preferably validation) may select the best model.

Parameters

inModel

oa::Module &

—

inOpt

oa::Optimizer &

—

inStep

oa::U64

—

inMetric

oa::F64

—

inMetricName

const oa::String &

Default: = {}

Returns

oa::Status

The declared return value.