Documentation ¶
Index ¶
- Variables
- type ExportedModel
- func (*ExportedModel) Descriptor() ([]byte, []int)deprecated
- func (x *ExportedModel) GetAssetFileDefs() []*protobuf.AssetFileDef
- func (x *ExportedModel) GetCheckpointDir() string
- func (x *ExportedModel) GetFunctionAliases() map[string]string
- func (x *ExportedModel) GetGraphDef() *framework.GraphDef
- func (x *ExportedModel) GetInitNodeName() string
- func (x *ExportedModel) GetSaverDef() *protobuf.SaverDef
- func (*ExportedModel) ProtoMessage()
- func (x *ExportedModel) ProtoReflect() protoreflect.Message
- func (x *ExportedModel) Reset()
- func (x *ExportedModel) String() string
- type FreezeAllVariables
- func (*FreezeAllVariables) Descriptor() ([]byte, []int)deprecated
- func (x *FreezeAllVariables) GetEnabled() bool
- func (*FreezeAllVariables) ProtoMessage()
- func (x *FreezeAllVariables) ProtoReflect() protoreflect.Message
- func (x *FreezeAllVariables) Reset()
- func (x *FreezeAllVariables) String() string
- type OpSet
- type QuantizationMethod
- func (*QuantizationMethod) Descriptor() ([]byte, []int)deprecated
- func (x *QuantizationMethod) GetExperimentalMethod() QuantizationMethod_ExperimentalMethod
- func (x *QuantizationMethod) GetMethod() QuantizationMethod_Method
- func (m *QuantizationMethod) GetMethodOneof() isQuantizationMethod_MethodOneof
- func (*QuantizationMethod) ProtoMessage()
- func (x *QuantizationMethod) ProtoReflect() protoreflect.Message
- func (x *QuantizationMethod) Reset()
- func (x *QuantizationMethod) String() string
- type QuantizationMethod_ExperimentalMethod
- func (QuantizationMethod_ExperimentalMethod) Descriptor() protoreflect.EnumDescriptor
- func (x QuantizationMethod_ExperimentalMethod) Enum() *QuantizationMethod_ExperimentalMethod
- func (QuantizationMethod_ExperimentalMethod) EnumDescriptor() ([]byte, []int)deprecated
- func (x QuantizationMethod_ExperimentalMethod) Number() protoreflect.EnumNumber
- func (x QuantizationMethod_ExperimentalMethod) String() string
- func (QuantizationMethod_ExperimentalMethod) Type() protoreflect.EnumType
- type QuantizationMethod_ExperimentalMethod_
- type QuantizationMethod_Method
- func (QuantizationMethod_Method) Descriptor() protoreflect.EnumDescriptor
- func (x QuantizationMethod_Method) Enum() *QuantizationMethod_Method
- func (QuantizationMethod_Method) EnumDescriptor() ([]byte, []int)deprecated
- func (x QuantizationMethod_Method) Number() protoreflect.EnumNumber
- func (x QuantizationMethod_Method) String() string
- func (QuantizationMethod_Method) Type() protoreflect.EnumType
- type QuantizationMethod_Method_
- type QuantizationOptions
- func (*QuantizationOptions) Descriptor() ([]byte, []int)deprecated
- func (x *QuantizationOptions) GetEnableLegacyWeightOnly() bool
- func (x *QuantizationOptions) GetEnablePerChannelQuantization() bool
- func (x *QuantizationOptions) GetEnableTwoInputTensors() bool
- func (x *QuantizationOptions) GetExperimentalEnableTpuModelSupport() bool
- func (x *QuantizationOptions) GetForceGraphModeCalibration() bool
- func (x *QuantizationOptions) GetFreezeAllVariables() *FreezeAllVariables
- func (x *QuantizationOptions) GetMinNumElementsForWeights() int64
- func (x *QuantizationOptions) GetOpSet() OpSet
- func (x *QuantizationOptions) GetQuantizationMethod() *QuantizationMethod
- func (x *QuantizationOptions) GetQuantizationPrecision() QuantizationPrecision
- func (x *QuantizationOptions) GetUnitWiseQuantizationPrecision() []*UnitWiseQuantizationPrecision
- func (*QuantizationOptions) ProtoMessage()
- func (x *QuantizationOptions) ProtoReflect() protoreflect.Message
- func (x *QuantizationOptions) Reset()
- func (x *QuantizationOptions) String() string
- type QuantizationPrecision
- func (QuantizationPrecision) Descriptor() protoreflect.EnumDescriptor
- func (x QuantizationPrecision) Enum() *QuantizationPrecision
- func (QuantizationPrecision) EnumDescriptor() ([]byte, []int)deprecated
- func (x QuantizationPrecision) Number() protoreflect.EnumNumber
- func (x QuantizationPrecision) String() string
- func (QuantizationPrecision) Type() protoreflect.EnumType
- type UnitWiseQuantizationPrecision
- func (*UnitWiseQuantizationPrecision) Descriptor() ([]byte, []int)deprecated
- func (x *UnitWiseQuantizationPrecision) GetFuncName() string
- func (x *UnitWiseQuantizationPrecision) GetQuantizationPrecision() QuantizationPrecision
- func (x *UnitWiseQuantizationPrecision) GetUnitName() string
- func (x *UnitWiseQuantizationPrecision) GetUnitType() UnitWiseQuantizationPrecision_UnitType
- func (*UnitWiseQuantizationPrecision) ProtoMessage()
- func (x *UnitWiseQuantizationPrecision) ProtoReflect() protoreflect.Message
- func (x *UnitWiseQuantizationPrecision) Reset()
- func (x *UnitWiseQuantizationPrecision) String() string
- type UnitWiseQuantizationPrecision_UnitType
- func (UnitWiseQuantizationPrecision_UnitType) Descriptor() protoreflect.EnumDescriptor
- func (x UnitWiseQuantizationPrecision_UnitType) Enum() *UnitWiseQuantizationPrecision_UnitType
- func (UnitWiseQuantizationPrecision_UnitType) EnumDescriptor() ([]byte, []int)deprecated
- func (x UnitWiseQuantizationPrecision_UnitType) Number() protoreflect.EnumNumber
- func (x UnitWiseQuantizationPrecision_UnitType) String() string
- func (UnitWiseQuantizationPrecision_UnitType) Type() protoreflect.EnumType
Constants ¶
This section is empty.
Variables ¶
var ( QuantizationPrecision_name = map[int32]string{ 0: "PRECISION_UNSPECIFIED", 1: "PRECISION_FULL", 2: "PRECISION_W4A4", 3: "PRECISION_W4A8", 4: "PRECISION_W8A8", } QuantizationPrecision_value = map[string]int32{ "PRECISION_UNSPECIFIED": 0, "PRECISION_FULL": 1, "PRECISION_W4A4": 2, "PRECISION_W4A8": 3, "PRECISION_W8A8": 4, } )
Enum value maps for QuantizationPrecision.
var ( OpSet_name = map[int32]string{ 0: "OP_SET_UNSPECIFIED", 1: "TF", 2: "XLA", 3: "UNIFORM_QUANTIZED", } OpSet_value = map[string]int32{ "OP_SET_UNSPECIFIED": 0, "TF": 1, "XLA": 2, "UNIFORM_QUANTIZED": 3, } )
Enum value maps for OpSet.
var ( QuantizationMethod_Method_name = map[int32]string{ 0: "METHOD_UNSPECIFIED", } QuantizationMethod_Method_value = map[string]int32{ "METHOD_UNSPECIFIED": 0, } )
Enum value maps for QuantizationMethod_Method.
var ( QuantizationMethod_ExperimentalMethod_name = map[int32]string{ 0: "EXPERIMENTAL_METHOD_UNSPECIFIED", 1: "STATIC_RANGE", 2: "DYNAMIC_RANGE", 3: "WEIGHT_ONLY", } QuantizationMethod_ExperimentalMethod_value = map[string]int32{ "EXPERIMENTAL_METHOD_UNSPECIFIED": 0, "STATIC_RANGE": 1, "DYNAMIC_RANGE": 2, "WEIGHT_ONLY": 3, } )
Enum value maps for QuantizationMethod_ExperimentalMethod.
var ( UnitWiseQuantizationPrecision_UnitType_name = map[int32]string{ 0: "UNIT_UNSPECIFIED", 1: "UNIT_NODE", 2: "UNIT_OP", } UnitWiseQuantizationPrecision_UnitType_value = map[string]int32{ "UNIT_UNSPECIFIED": 0, "UNIT_NODE": 1, "UNIT_OP": 2, } )
Enum value maps for UnitWiseQuantizationPrecision_UnitType.
var File_tensorflow_compiler_mlir_quantization_tensorflow_exported_model_proto protoreflect.FileDescriptor
var File_tensorflow_compiler_mlir_quantization_tensorflow_quantization_options_proto protoreflect.FileDescriptor
Functions ¶
This section is empty.
Types ¶
type ExportedModel ¶ added in v2.12.0
type ExportedModel struct { GraphDef *framework.GraphDef `protobuf:"bytes,1,opt,name=graph_def,json=graphDef,proto3" json:"graph_def,omitempty"` // Name of the initialization node (TF Operation) used for initializing // resources like hash tables upon loading. InitNodeName string `protobuf:"bytes,2,opt,name=init_node_name,json=initNodeName,proto3" json:"init_node_name,omitempty"` // Path to the directory where checkpoint files are saved. This directoy is // not expected to be persistent (usually a temporary directory). When // fetching the restore op (see `restore_node_name`), this value is provided // to the "file_prefix" tensor to identify the checkpoint directory. CheckpointDir string `protobuf:"bytes,5,opt,name=checkpoint_dir,json=checkpointDir,proto3" json:"checkpoint_dir,omitempty"` // Function name -> function alias mapping. This associates the quantized // functions to the original functions' aliases. This information will be used // to populate `MetaInfoDef`s `function_aliases` when the quantized model is // exported to the saved model. This field is usually only populated for the // TF2 models. FunctionAliases map[string]string `` /* 194-byte string literal not displayed */ // Holds information about the asset files used for the model. It essentially // associates asset file names with the tensors to which the asset file names // should be fed. AssetFileDefs []*protobuf.AssetFileDef `protobuf:"bytes,8,rep,name=asset_file_defs,json=assetFileDefs,proto3" json:"asset_file_defs,omitempty"` // SaverDef including the information required for saving and restoring // variables. This field is not set if there are no variables in the exported // graph. If set, the fields `version`, `filename_tensor_name`, // `restore_op_name` and `save_tensor_name` are populated. SaverDef *protobuf.SaverDef `protobuf:"bytes,10,opt,name=saver_def,json=saverDef,proto3" json:"saver_def,omitempty"` // contains filtered or unexported fields }
Represents an exported TensorFlow model. It consists of a GraphDef and extra metadata required for building a SavedModel. This message is primarily used to "export" the model produced from various quantization passes in c++ to Python layer. Next ID: 11
func (*ExportedModel) Descriptor
deprecated
added in
v2.12.0
func (*ExportedModel) Descriptor() ([]byte, []int)
Deprecated: Use ExportedModel.ProtoReflect.Descriptor instead.
func (*ExportedModel) GetAssetFileDefs ¶ added in v2.13.0
func (x *ExportedModel) GetAssetFileDefs() []*protobuf.AssetFileDef
func (*ExportedModel) GetCheckpointDir ¶ added in v2.12.0
func (x *ExportedModel) GetCheckpointDir() string
func (*ExportedModel) GetFunctionAliases ¶ added in v2.12.0
func (x *ExportedModel) GetFunctionAliases() map[string]string
func (*ExportedModel) GetGraphDef ¶ added in v2.12.0
func (x *ExportedModel) GetGraphDef() *framework.GraphDef
func (*ExportedModel) GetInitNodeName ¶ added in v2.12.0
func (x *ExportedModel) GetInitNodeName() string
func (*ExportedModel) GetSaverDef ¶ added in v2.13.0
func (x *ExportedModel) GetSaverDef() *protobuf.SaverDef
func (*ExportedModel) ProtoMessage ¶ added in v2.12.0
func (*ExportedModel) ProtoMessage()
func (*ExportedModel) ProtoReflect ¶ added in v2.12.0
func (x *ExportedModel) ProtoReflect() protoreflect.Message
func (*ExportedModel) Reset ¶ added in v2.12.0
func (x *ExportedModel) Reset()
func (*ExportedModel) String ¶ added in v2.12.0
func (x *ExportedModel) String() string
type FreezeAllVariables ¶ added in v2.12.0
type FreezeAllVariables struct { // Setting this to true freezes all variables to constants during // quantization. Setting this to `false` is an experimental feature and does // not have stability guarantees. Enabled bool `protobuf:"varint,1,opt,name=enabled,proto3" json:"enabled,omitempty"` // contains filtered or unexported fields }
Configurations for variable freezing during quantization passes. NEXT ID: 2
func (*FreezeAllVariables) Descriptor
deprecated
added in
v2.12.0
func (*FreezeAllVariables) Descriptor() ([]byte, []int)
Deprecated: Use FreezeAllVariables.ProtoReflect.Descriptor instead.
func (*FreezeAllVariables) GetEnabled ¶ added in v2.12.0
func (x *FreezeAllVariables) GetEnabled() bool
func (*FreezeAllVariables) ProtoMessage ¶ added in v2.12.0
func (*FreezeAllVariables) ProtoMessage()
func (*FreezeAllVariables) ProtoReflect ¶ added in v2.12.0
func (x *FreezeAllVariables) ProtoReflect() protoreflect.Message
func (*FreezeAllVariables) Reset ¶ added in v2.12.0
func (x *FreezeAllVariables) Reset()
func (*FreezeAllVariables) String ¶ added in v2.12.0
func (x *FreezeAllVariables) String() string
type OpSet ¶
type OpSet int32
List of supported opsets to deploy the quantized model. The quantized model contains different set of ops depending on the opset. NEXT ID: 4
const ( OpSet_OP_SET_UNSPECIFIED OpSet = 0 // go/do-include-enum-unspecified // Uses TF ops that mimic quantization behavior. Used when the corresponding // integer op is not yet present. OpSet_TF OpSet = 1 // Uses TF XLA ops OpSet_XLA OpSet = 2 // Uses TF Uniform Quantized ops OpSet_UNIFORM_QUANTIZED OpSet = 3 )
func (OpSet) Descriptor ¶
func (OpSet) Descriptor() protoreflect.EnumDescriptor
func (OpSet) EnumDescriptor
deprecated
func (OpSet) Number ¶
func (x OpSet) Number() protoreflect.EnumNumber
func (OpSet) Type ¶
func (OpSet) Type() protoreflect.EnumType
type QuantizationMethod ¶
type QuantizationMethod struct { // Quantization method is either exprimental or non-experimental method. // // Types that are assignable to MethodOneof: // // *QuantizationMethod_Method_ // *QuantizationMethod_ExperimentalMethod_ MethodOneof isQuantizationMethod_MethodOneof `protobuf_oneof:"method_oneof"` // contains filtered or unexported fields }
Model quantization method for optimization.
Various techniques for model quantization are defined within this message along with a field that specifies a method to be used for a particular quantization request. NEXT ID: 3
func (*QuantizationMethod) Descriptor
deprecated
func (*QuantizationMethod) Descriptor() ([]byte, []int)
Deprecated: Use QuantizationMethod.ProtoReflect.Descriptor instead.
func (*QuantizationMethod) GetExperimentalMethod ¶
func (x *QuantizationMethod) GetExperimentalMethod() QuantizationMethod_ExperimentalMethod
func (*QuantizationMethod) GetMethod ¶
func (x *QuantizationMethod) GetMethod() QuantizationMethod_Method
func (*QuantizationMethod) GetMethodOneof ¶
func (m *QuantizationMethod) GetMethodOneof() isQuantizationMethod_MethodOneof
func (*QuantizationMethod) ProtoMessage ¶
func (*QuantizationMethod) ProtoMessage()
func (*QuantizationMethod) ProtoReflect ¶
func (x *QuantizationMethod) ProtoReflect() protoreflect.Message
func (*QuantizationMethod) Reset ¶
func (x *QuantizationMethod) Reset()
func (*QuantizationMethod) String ¶
func (x *QuantizationMethod) String() string
type QuantizationMethod_ExperimentalMethod ¶
type QuantizationMethod_ExperimentalMethod int32
Experimental quantization methods. These methods are either not implemented or provided with an unstable behavior.
const ( // This should never be used. Using this will generally result in an error. QuantizationMethod_EXPERIMENTAL_METHOD_UNSPECIFIED QuantizationMethod_ExperimentalMethod = 0 // go/do-include-enum-unspecified // Static range quantization. Quantized tensor values' ranges are statically // determined. QuantizationMethod_STATIC_RANGE QuantizationMethod_ExperimentalMethod = 1 // Dynamic range quantization. Quantized tensor values' ranges are // determined in the graph executions. The weights are quantized during // conversion. QuantizationMethod_DYNAMIC_RANGE QuantizationMethod_ExperimentalMethod = 2 // Weight-only quantization. Only weights are quantized during conversion. QuantizationMethod_WEIGHT_ONLY QuantizationMethod_ExperimentalMethod = 3 )
func (QuantizationMethod_ExperimentalMethod) Descriptor ¶
func (QuantizationMethod_ExperimentalMethod) Descriptor() protoreflect.EnumDescriptor
func (QuantizationMethod_ExperimentalMethod) EnumDescriptor
deprecated
func (QuantizationMethod_ExperimentalMethod) EnumDescriptor() ([]byte, []int)
Deprecated: Use QuantizationMethod_ExperimentalMethod.Descriptor instead.
func (QuantizationMethod_ExperimentalMethod) Number ¶
func (x QuantizationMethod_ExperimentalMethod) Number() protoreflect.EnumNumber
func (QuantizationMethod_ExperimentalMethod) String ¶
func (x QuantizationMethod_ExperimentalMethod) String() string
func (QuantizationMethod_ExperimentalMethod) Type ¶
func (QuantizationMethod_ExperimentalMethod) Type() protoreflect.EnumType
type QuantizationMethod_ExperimentalMethod_ ¶
type QuantizationMethod_ExperimentalMethod_ struct {
ExperimentalMethod QuantizationMethod_ExperimentalMethod `` /* 152-byte string literal not displayed */
}
type QuantizationMethod_Method ¶
type QuantizationMethod_Method int32
Quantization methods that are supported as a stable API.
const ( // This should never be used. Using this will generally result in an error. QuantizationMethod_METHOD_UNSPECIFIED QuantizationMethod_Method = 0 // go/do-include-enum-unspecified )
func (QuantizationMethod_Method) Descriptor ¶
func (QuantizationMethod_Method) Descriptor() protoreflect.EnumDescriptor
func (QuantizationMethod_Method) Enum ¶
func (x QuantizationMethod_Method) Enum() *QuantizationMethod_Method
func (QuantizationMethod_Method) EnumDescriptor
deprecated
func (QuantizationMethod_Method) EnumDescriptor() ([]byte, []int)
Deprecated: Use QuantizationMethod_Method.Descriptor instead.
func (QuantizationMethod_Method) Number ¶
func (x QuantizationMethod_Method) Number() protoreflect.EnumNumber
func (QuantizationMethod_Method) String ¶
func (x QuantizationMethod_Method) String() string
func (QuantizationMethod_Method) Type ¶
func (QuantizationMethod_Method) Type() protoreflect.EnumType
type QuantizationMethod_Method_ ¶
type QuantizationMethod_Method_ struct {
Method QuantizationMethod_Method `protobuf:"varint,1,opt,name=method,proto3,enum=tensorflow.quantization.QuantizationMethod_Method,oneof"`
}
type QuantizationOptions ¶
type QuantizationOptions struct { // The default quantization configuration for the model. If the below // unit-wise configuration does not exist, we use this default quantization // configuration for the entire model. If the below unit-wise configuration // exists, this default one will become the quantization configuration for // units that are not specified in unit-wise configurations. QuantizationMethod *QuantizationMethod `protobuf:"bytes,1,opt,name=quantization_method,json=quantizationMethod,proto3" json:"quantization_method,omitempty"` OpSet OpSet `protobuf:"varint,2,opt,name=op_set,json=opSet,proto3,enum=tensorflow.quantization.OpSet" json:"op_set,omitempty"` // If not specified, it defaults to `XLA`. QuantizationPrecision QuantizationPrecision `` /* 176-byte string literal not displayed */ // Quantization precision for each unit. Units can become either // nodes or ops, and the mixture of those different units are allowed. // If there are conflicts or ambiguity in this unit-wise precision, our // quantizer will raise an error. UnitWiseQuantizationPrecision []*UnitWiseQuantizationPrecision `` /* 152-byte string literal not displayed */ // Minimum number of weight elements to apply quantization. Currently only // supported for Post-training Dynamic Range Quantization. By default, it is // set to 1024. To disable this, set the value to -1 explicitly. MinNumElementsForWeights int64 `` /* 140-byte string literal not displayed */ // When set to `true`, freezes all variables in the model into constants. // When set to `false` the model's large constants are converted to variables. // Setting this to `false` is an experimental feature and quantization may // fail. To quantize models larger than 2 GiB, this should be set to `false`. // If not set, it defaults to `true`. FreezeAllVariables *FreezeAllVariables `protobuf:"bytes,6,opt,name=freeze_all_variables,json=freezeAllVariables,proto3" json:"freeze_all_variables,omitempty"` // Enables chnanel-wise quantizaiton. By default, channel-wise quantization is // not applied regardless of the op support. Currently, it is supported for // Uniform Quantized opset only. EnablePerChannelQuantization bool `` /* 150-byte string literal not displayed */ // Enables two inputs of an operation to be both tensors. // Currently supports MatMul and BatchMatMul ops for XLA. // TODO(b/263528090): Check the condition when this feature is beneficial. EnableTwoInputTensors bool `` /* 129-byte string literal not displayed */ // Supports TPU model quantization. If the target model for the quantization // is already converted for TPU, this flag may be helpful. Note that this // feature may be unstable as it is under the experimental stage. ExperimentalEnableTpuModelSupport bool `` /* 167-byte string literal not displayed */ // Produces legacy weight-only graph where the qconst op(containing quantized // values) is followed by a dequantization op. EnableLegacyWeightOnly bool `` /* 133-byte string literal not displayed */ // If set to true, it forces calibration in graph model instead of eager mode // when the context is in eager mode. ForceGraphModeCalibration bool `` /* 142-byte string literal not displayed */ // contains filtered or unexported fields }
Defines various options to specify and control the behavior of the quantizer. It consists of 1) Model-wise quantization configuration as a default configuration. If it is None, the default configuration is "do not quantize the model". 2) A set of supported operations. 3) Unit wise quantization precision. 4) Target hardware name. NEXT ID: 12
func (*QuantizationOptions) Descriptor
deprecated
func (*QuantizationOptions) Descriptor() ([]byte, []int)
Deprecated: Use QuantizationOptions.ProtoReflect.Descriptor instead.
func (*QuantizationOptions) GetEnableLegacyWeightOnly ¶ added in v2.13.0
func (x *QuantizationOptions) GetEnableLegacyWeightOnly() bool
func (*QuantizationOptions) GetEnablePerChannelQuantization ¶ added in v2.12.0
func (x *QuantizationOptions) GetEnablePerChannelQuantization() bool
func (*QuantizationOptions) GetEnableTwoInputTensors ¶ added in v2.12.0
func (x *QuantizationOptions) GetEnableTwoInputTensors() bool
func (*QuantizationOptions) GetExperimentalEnableTpuModelSupport ¶ added in v2.13.0
func (x *QuantizationOptions) GetExperimentalEnableTpuModelSupport() bool
func (*QuantizationOptions) GetForceGraphModeCalibration ¶ added in v2.13.0
func (x *QuantizationOptions) GetForceGraphModeCalibration() bool
func (*QuantizationOptions) GetFreezeAllVariables ¶ added in v2.12.0
func (x *QuantizationOptions) GetFreezeAllVariables() *FreezeAllVariables
func (*QuantizationOptions) GetMinNumElementsForWeights ¶ added in v2.11.0
func (x *QuantizationOptions) GetMinNumElementsForWeights() int64
func (*QuantizationOptions) GetOpSet ¶
func (x *QuantizationOptions) GetOpSet() OpSet
func (*QuantizationOptions) GetQuantizationMethod ¶
func (x *QuantizationOptions) GetQuantizationMethod() *QuantizationMethod
func (*QuantizationOptions) GetQuantizationPrecision ¶ added in v2.11.0
func (x *QuantizationOptions) GetQuantizationPrecision() QuantizationPrecision
func (*QuantizationOptions) GetUnitWiseQuantizationPrecision ¶ added in v2.11.0
func (x *QuantizationOptions) GetUnitWiseQuantizationPrecision() []*UnitWiseQuantizationPrecision
func (*QuantizationOptions) ProtoMessage ¶
func (*QuantizationOptions) ProtoMessage()
func (*QuantizationOptions) ProtoReflect ¶
func (x *QuantizationOptions) ProtoReflect() protoreflect.Message
func (*QuantizationOptions) Reset ¶
func (x *QuantizationOptions) Reset()
func (*QuantizationOptions) String ¶
func (x *QuantizationOptions) String() string
type QuantizationPrecision ¶ added in v2.11.0
type QuantizationPrecision int32
Quantization precisions. If the specified quantization precision is not available, our quantizer needs to raise an error.
const ( QuantizationPrecision_PRECISION_UNSPECIFIED QuantizationPrecision = 0 // Full Precision (Do not quantize) QuantizationPrecision_PRECISION_FULL QuantizationPrecision = 1 // Weight 4 bit and activation 4 bit quantization QuantizationPrecision_PRECISION_W4A4 QuantizationPrecision = 2 // Weight 4 bit and activation 8 bit quantization QuantizationPrecision_PRECISION_W4A8 QuantizationPrecision = 3 // Weight 8 bit and activation 8 bit quantization QuantizationPrecision_PRECISION_W8A8 QuantizationPrecision = 4 )
func (QuantizationPrecision) Descriptor ¶ added in v2.11.0
func (QuantizationPrecision) Descriptor() protoreflect.EnumDescriptor
func (QuantizationPrecision) Enum ¶ added in v2.11.0
func (x QuantizationPrecision) Enum() *QuantizationPrecision
func (QuantizationPrecision) EnumDescriptor
deprecated
added in
v2.11.0
func (QuantizationPrecision) EnumDescriptor() ([]byte, []int)
Deprecated: Use QuantizationPrecision.Descriptor instead.
func (QuantizationPrecision) Number ¶ added in v2.11.0
func (x QuantizationPrecision) Number() protoreflect.EnumNumber
func (QuantizationPrecision) String ¶ added in v2.11.0
func (x QuantizationPrecision) String() string
func (QuantizationPrecision) Type ¶ added in v2.11.0
func (QuantizationPrecision) Type() protoreflect.EnumType
type UnitWiseQuantizationPrecision ¶ added in v2.11.0
type UnitWiseQuantizationPrecision struct { // Available quantization unit. Currently node-wise and op-wise are // available quantization units. UnitType UnitWiseQuantizationPrecision_UnitType `` /* 154-byte string literal not displayed */ // Uniqueness isn't guaranteed across SavedModels but within each function // def's level, uniqueness is guaranteed. Updated // the configuration interfaces to reflect such circumstances. // If users do not need to guarantee uniqueness func_name can be omitted. FuncName string `protobuf:"bytes,2,opt,name=func_name,json=funcName,proto3" json:"func_name,omitempty"` UnitName string `protobuf:"bytes,3,opt,name=unit_name,json=unitName,proto3" json:"unit_name,omitempty"` // Quantization option information for the current unit. // TODO(b/241322587): Support specifying quantization method for each unit of // TF GraphDef. QuantizationPrecision QuantizationPrecision `` /* 176-byte string literal not displayed */ // contains filtered or unexported fields }
Unit (either nodes or ops at this moment) wise quantization method for mixed bit precision quantization. It contains the name of the unit, the granularity of the unit, and the quantization method for each unit. NEXT ID: 6
func (*UnitWiseQuantizationPrecision) Descriptor
deprecated
added in
v2.11.0
func (*UnitWiseQuantizationPrecision) Descriptor() ([]byte, []int)
Deprecated: Use UnitWiseQuantizationPrecision.ProtoReflect.Descriptor instead.
func (*UnitWiseQuantizationPrecision) GetFuncName ¶ added in v2.11.0
func (x *UnitWiseQuantizationPrecision) GetFuncName() string
func (*UnitWiseQuantizationPrecision) GetQuantizationPrecision ¶ added in v2.11.0
func (x *UnitWiseQuantizationPrecision) GetQuantizationPrecision() QuantizationPrecision
func (*UnitWiseQuantizationPrecision) GetUnitName ¶ added in v2.11.0
func (x *UnitWiseQuantizationPrecision) GetUnitName() string
func (*UnitWiseQuantizationPrecision) GetUnitType ¶ added in v2.11.0
func (x *UnitWiseQuantizationPrecision) GetUnitType() UnitWiseQuantizationPrecision_UnitType
func (*UnitWiseQuantizationPrecision) ProtoMessage ¶ added in v2.11.0
func (*UnitWiseQuantizationPrecision) ProtoMessage()
func (*UnitWiseQuantizationPrecision) ProtoReflect ¶ added in v2.11.0
func (x *UnitWiseQuantizationPrecision) ProtoReflect() protoreflect.Message
func (*UnitWiseQuantizationPrecision) Reset ¶ added in v2.11.0
func (x *UnitWiseQuantizationPrecision) Reset()
func (*UnitWiseQuantizationPrecision) String ¶ added in v2.11.0
func (x *UnitWiseQuantizationPrecision) String() string
type UnitWiseQuantizationPrecision_UnitType ¶ added in v2.11.0
type UnitWiseQuantizationPrecision_UnitType int32
Quantization unit granularity. NEXT ID: 3
const ( // This should never be used. Using this will generally result in an error. UnitWiseQuantizationPrecision_UNIT_UNSPECIFIED UnitWiseQuantizationPrecision_UnitType = 0 UnitWiseQuantizationPrecision_UNIT_NODE UnitWiseQuantizationPrecision_UnitType = 1 UnitWiseQuantizationPrecision_UNIT_OP UnitWiseQuantizationPrecision_UnitType = 2 )
func (UnitWiseQuantizationPrecision_UnitType) Descriptor ¶ added in v2.11.0
func (UnitWiseQuantizationPrecision_UnitType) Descriptor() protoreflect.EnumDescriptor
func (UnitWiseQuantizationPrecision_UnitType) EnumDescriptor
deprecated
added in
v2.11.0
func (UnitWiseQuantizationPrecision_UnitType) EnumDescriptor() ([]byte, []int)
Deprecated: Use UnitWiseQuantizationPrecision_UnitType.Descriptor instead.
func (UnitWiseQuantizationPrecision_UnitType) Number ¶ added in v2.11.0
func (x UnitWiseQuantizationPrecision_UnitType) Number() protoreflect.EnumNumber
func (UnitWiseQuantizationPrecision_UnitType) String ¶ added in v2.11.0
func (x UnitWiseQuantizationPrecision_UnitType) String() string
func (UnitWiseQuantizationPrecision_UnitType) Type ¶ added in v2.11.0
func (UnitWiseQuantizationPrecision_UnitType) Type() protoreflect.EnumType