CustomValidationConfig

AI-generated Key Takeaways

• CustomValidationConfig provides custom validation data for benchmarking TensorFlow Lite models.

• It allows defining custom accuracy validation rules using AccuracyValidator and provides options to specify golden inputs and outputs.

• CustomValidationConfig offers pre-defined validators like BYTE_MATCHING_VALIDATOR and SKIP_VALIDATION for common use cases.

• Developers can configure golden config and batch size for validation, with DEFAULT_GOLDEN_CONFIG available as a fallback.

• The class provides getter methods to access the configured accuracy validator, batch size, golden config, golden inputs, and golden outputs.

public class CustomValidationConfig extends ValidationConfig

Provides custom validation data for benchmarking.

Nested Class Summary

interface	CustomValidationConfig.AccuracyValidator		Defines custom accuracy validation rule.
class	CustomValidationConfig.Builder		Builder class.

Inherited Constant Summary

From class com.google.android.gms.tflite.acceleration.ValidationConfig

long

DEFAULT_INFERENCE_TIMEOUT_MILLIS

The default test timeout in milli-seconds.

Field Summary

public static final CustomValidationConfig.AccuracyValidator	BYTE_MATCHING_VALIDATOR	Validates model outputs byte-by-byte.
public static final AccelerationConfig	DEFAULT_GOLDEN_CONFIG	Default config used to calculate golden output when golden is not provided.
public static final CustomValidationConfig.AccuracyValidator	SKIP_VALIDATION	No-op validator to skip accuracy validation.

Public Method Summary

CustomValidationConfig.AccuracyValidator	accuracyValidator() Gets accuracy validator value.
int	batchSize() Gets batch size value.
AccelerationConfig	goldenConfig() Gets golden config value.
Object[]	goldenInputs() Gets golden inputs value.
ByteBuffer[]	goldenOutputs() Gets golden outputs value.

Inherited Method Summary

From class com.google.android.gms.tflite.acceleration.ValidationConfig

long	inferenceTimeoutMillis() Returns test timeout in milli-seconds.

From class java.lang.Object

Object	clone()
boolean	equals(Object arg0)
void	finalize()
final Class<?>	getClass()
int	hashCode()
final void	notify()
final void	notifyAll()
String	toString()
final void	wait(long arg0, int arg1)
final void	wait(long arg0)
final void	wait()

Fields

public static final CustomValidationConfig.AccuracyValidator BYTE_MATCHING_VALIDATOR

Validates model outputs byte-by-byte. Output tensors are compared using ByteBuffer.equals(Object) function. This validator will return false if either golden output or benchmark output is null, or their dimensions do not match.

public static final AccelerationConfig DEFAULT_GOLDEN_CONFIG

Default config used to calculate golden output when golden is not provided.

public static final CustomValidationConfig.AccuracyValidator SKIP_VALIDATION

No-op validator to skip accuracy validation.

Public Methods

public CustomValidationConfig.AccuracyValidator accuracyValidator ()

Gets accuracy validator value.

public int batchSize ()

Gets batch size value.

public AccelerationConfig goldenConfig ()

Gets golden config value.

public Object[] goldenInputs ()

Gets golden inputs value.

public ByteBuffer[] goldenOutputs ()

Gets golden outputs value.