tflite_bindings library

Classes

TFLGpuDelegateOptions

Creates a new delegate instance that need to be destroyed with DeleteFlowDelegate when delegate is no longer used by tflite.

TFLGpuDelegateWaitType

tensorflow#metal_delegate

TfLiteCoreMlDelegateEnabledDevices

tensorflow#coreml_delegate

TfLiteCoreMlDelegateOptions

TfLiteGpuDelegateOptionsV2

IMPORTANT: Always use TfLiteGpuDelegateOptionsV2Default() method to create new instance of TfLiteGpuDelegateOptionsV2, otherwise every new added option may break inference.

TfLiteGpuExperimentalFlags

Used to toggle experimental flags used in the delegate. Note that this is a bitmask, so the values should be 1, 2, 4, 8, ...etc.

TfLiteGpuInferencePriority

TfLiteGpuInferenceUsage

tensorflow#gpu_delegate Encapsulated compilation/runtime tradeoffs.

TfLiteQuantizationParams

Legacy. Will be deprecated in favor of TfLiteAffineQuantization. If per-layer quantization is specified this field will still be populated in addition to TfLiteAffineQuantization. Parameters for asymmetric quantization. Quantized values can be converted back to float using: real_value = scale * (quantized_value - zero_point)

TfLiteStatus

tensorflow#c_api_types Note that new error status values may be added in future in order to indicate more fine-grained internal states, therefore, applications should not rely on status values being members of the enum.

TfLiteType

Types supported by tensor

TfLiteXNNPackDelegateOptions

TfLiteXNNPackDelegateWeightsCache

Constants

TFLITE_XNNPACK_DELEGATE_FLAG_FORCE_FP16 → const int

Force FP16 inference for FP32 operators.

TFLITE_XNNPACK_DELEGATE_FLAG_QS8 → const int

tensorflow#xnnpack_delegate Enable XNNPACK acceleration for signed quantized 8-bit inference. This includes operators with channel-wise quantized weights.

TFLITE_XNNPACK_DELEGATE_FLAG_QU8 → const int

Enable XNNPACK acceleration for unsigned quantized 8-bit inference.

Properties

TFLGpuDelegateCreate → Pointer<TfLiteDelegate> Function(Pointer<TFLGpuDelegateOptions>? options)

Creates a new delegate instance that need to be destroyed with TFLDeleteTfLiteGpuDelegate when delegate is no longer used by TFLite. When options is set to nullptr, the following default values are used: .precision_loss_allowed = false, .wait_type = kPassive,

final

TFLGpuDelegateDelete → void Function(Pointer<TfLiteDelegate>)

Destroys a delegate created with TFLGpuDelegateCreate call.

final

TFLGpuDelegateOptionsDefault → TFLGpuDelegateOptions Function()

Populates TFLGpuDelegateOptions as follows: allow_precision_loss = false; wait_type = TFLGpuDelegateWaitType::TFLGpuDelegateWaitTypePassive; enable_quantization = true;

final

TfLiteCoreMlDelegateCreate → Pointer<TfLiteDelegate> Function(Pointer<TfLiteCoreMlDelegateOptions> options)

final

TfLiteCoreMlDelegateDelete → void Function(Pointer<TfLiteDelegate> delegate)

final

TfLiteGpuDelegateOptionsV2Default → TfLiteGpuDelegateOptionsV2 Function()

Populates TfLiteGpuDelegateOptionsV2 as follows: is_precision_loss_allowed = false inference_preference = TFLITE_GPU_INFERENCE_PREFERENCE_FAST_SINGLE_ANSWER priority1 = TFLITE_GPU_INFERENCE_PRIORITY_MAX_PRECISION priority2 = TFLITE_GPU_INFERENCE_PRIORITY_AUTO priority3 = TFLITE_GPU_INFERENCE_PRIORITY_AUTO experimental_flags = TFLITE_GPU_EXPERIMENTAL_FLAGS_ENABLE_QUANT max_delegated_partitions = 1

final

TfLiteGpuDelegateV2Create → Pointer<TfLiteDelegate> Function(Pointer<TfLiteGpuDelegateOptionsV2> options)

Creates a new delegate instance that need to be destroyed with TfLiteGpuDelegateV2Delete when delegate is no longer used by TFLite.

final

TfLiteGpuDelegateV2Delete → void Function(Pointer<TfLiteDelegate> delegate)

Destroys a delegate created with TfLiteGpuDelegateV2Create call.

final

TfLiteXNNPackDelegateCreate → Pointer<TfLiteDelegate> Function(Pointer<TfLiteXNNPackDelegateOptions> options)

Creates a new delegate instance that need to be destroyed with TfLiteXNNPackDelegateDelete when delegate is no longer used by TFLite. When options is set to nullptr, default values are used (see implementation of TfLiteXNNPackDelegateOptionsDefault in the .cc file for details).

final

TfLiteXNNPackDelegateDelete → void Function(Pointer<TfLiteDelegate> delegate)

Destroys a delegate created with TfLiteXNNPackDelegateCreate call.

final

TfLiteXNNPackDelegateOptionsDefault → TfLiteXNNPackDelegateOptions Function()

Returns a structure with the default XNNPack delegate options.

final

TfLiteXNNPackDelegateWeightsCacheCreate → Pointer<TfLiteXNNPackDelegateWeightsCache> Function()

Creates a new weights cache that can be shared with multiple delegate instances.

final

TfLiteXNNPackWeightsCacheDelete → void Function(Pointer<TfLiteXNNPackDelegateWeightsCache> cache)

Destroys a weights cache created with TfLiteXNNPackDelegateWeightsCacheCreate call.

final