Evaluator Class Reference

Model manager that shares native weights across threads. More...

#include <Evaluator.h>

Public Member Functions
	Evaluator (const std::string &modelPath, int numThreads, bool useGPU=false)
	Constructs an Evaluator manager.
std::unique_ptr< IEvaluator >	createThreadContext ()
	Creates a lightweight thread context from the shared model.
bool	isLoaded () const
bool	isMoE () const
bool	isMoECache () const

Private Attributes
std::string	modelPath_
std::string	nativeWeightsPath_
int	numThreads_
bool	isMoECache_
bool	useGPU_
std::shared_ptr< const EvalContextMoECacheSharedModel >	sharedMoECacheModel_
	Shared native model used by all thread contexts.

Detailed Description

Model manager that shares native weights across threads.

It is responsible for loading the neural network architecture (or falling back to simple evaluation) and instantiating thread-local IEvaluator contexts that share the read-only weights to save memory.

Constructor & Destructor Documentation

Evaluator()

Evaluator::Evaluator	(	const std::string &	modelPath,
		int	numThreads,
		bool	useGPU = false )

Constructs an Evaluator manager.

Parameters

modelPath	Path to the network weights file.
numThreads	Number of threads that will request contexts.
useGPU	Whether to accelerate evaluation using the GPU (if supported).

Here is the call graph for this function:

Member Function Documentation

createThreadContext()

std::unique_ptr< IEvaluator > Evaluator::createThreadContext

(

)

Creates a lightweight thread context from the shared model.

Returns

A unique pointer to an IEvaluator instance for a specific thread.

Here is the caller graph for this function:

isLoaded()

bool Evaluator::isLoaded ( ) const

inline

Here is the caller graph for this function:

isMoE()

bool Evaluator::isMoE ( ) const

inline

isMoECache()

bool Evaluator::isMoECache ( ) const

inline

Member Data Documentation

isMoECache_

bool Evaluator::isMoECache_

private

modelPath_

std::string Evaluator::modelPath_

private

nativeWeightsPath_

std::string Evaluator::nativeWeightsPath_

private

numThreads_

int Evaluator::numThreads_

private

sharedMoECacheModel_

std::shared_ptr<const EvalContextMoECacheSharedModel> Evaluator::sharedMoECacheModel_

private

Shared native model used by all thread contexts.

useGPU_

bool Evaluator::useGPU_

private

---

The documentation for this class was generated from the following files:

• engine/src/eval/Evaluator.h
• engine/src/eval/Evaluator.cpp