Documentation ¶
Index ¶
- Variables
- type BatchingParameters
- func (*BatchingParameters) Descriptor() ([]byte, []int)deprecated
- func (x *BatchingParameters) GetAllowedBatchSizes() []int64
- func (x *BatchingParameters) GetBatchTimeoutMicros() *wrapperspb.Int64Value
- func (x *BatchingParameters) GetEnableLargeBatchSplitting() *wrapperspb.BoolValue
- func (x *BatchingParameters) GetMaxBatchSize() *wrapperspb.Int64Value
- func (x *BatchingParameters) GetMaxEnqueuedBatches() *wrapperspb.Int64Value
- func (x *BatchingParameters) GetMaxExecutionBatchSize() *wrapperspb.Int64Value
- func (x *BatchingParameters) GetNumBatchThreads() *wrapperspb.Int64Value
- func (x *BatchingParameters) GetPadVariableLengthInputs() bool
- func (x *BatchingParameters) GetThreadPoolName() *wrapperspb.StringValue
- func (*BatchingParameters) ProtoMessage()
- func (x *BatchingParameters) ProtoReflect() protoreflect.Message
- func (x *BatchingParameters) Reset()
- func (x *BatchingParameters) String() string
- type ModelWarmupOptions
- func (*ModelWarmupOptions) Descriptor() ([]byte, []int)deprecated
- func (x *ModelWarmupOptions) GetEnableAllBatchSizesWarmup() bool
- func (x *ModelWarmupOptions) GetModelName() string
- func (x *ModelWarmupOptions) GetModelVersion() int64
- func (x *ModelWarmupOptions) GetNumModelWarmupThreads() *wrapperspb.Int32Value
- func (x *ModelWarmupOptions) GetNumRequestIterations() *wrapperspb.Int32Value
- func (*ModelWarmupOptions) ProtoMessage()
- func (x *ModelWarmupOptions) ProtoReflect() protoreflect.Message
- func (x *ModelWarmupOptions) Reset()
- func (x *ModelWarmupOptions) String() string
- type RemoteOpRemapConfig
- func (*RemoteOpRemapConfig) Descriptor() ([]byte, []int)deprecated
- func (x *RemoteOpRemapConfig) GetBackendModelNameToTargetAddressRemap() map[string]string
- func (x *RemoteOpRemapConfig) GetModelNameRemap() map[string]string
- func (x *RemoteOpRemapConfig) GetPropagateModelName() bool
- func (x *RemoteOpRemapConfig) GetPropagateVersion() bool
- func (x *RemoteOpRemapConfig) GetTargetAddressRemap() map[string]string
- func (*RemoteOpRemapConfig) ProtoMessage()
- func (x *RemoteOpRemapConfig) ProtoReflect() protoreflect.Message
- func (x *RemoteOpRemapConfig) Reset()
- func (x *RemoteOpRemapConfig) String() string
- type SavedModelBundleSourceAdapterConfig
- func (*SavedModelBundleSourceAdapterConfig) Descriptor() ([]byte, []int)deprecated
- func (x *SavedModelBundleSourceAdapterConfig) GetLegacyConfig() *SessionBundleConfig
- func (*SavedModelBundleSourceAdapterConfig) ProtoMessage()
- func (x *SavedModelBundleSourceAdapterConfig) ProtoReflect() protoreflect.Message
- func (x *SavedModelBundleSourceAdapterConfig) Reset()
- func (x *SavedModelBundleSourceAdapterConfig) String() string
- type SavedModelConfig
- func (*SavedModelConfig) Descriptor() ([]byte, []int)deprecated
- func (x *SavedModelConfig) GetCritical() bool
- func (x *SavedModelConfig) GetSessionOverrides() *SessionOverrides
- func (x *SavedModelConfig) GetTfrtRuntimeConfig() *graph_executor_go_proto.RuntimeConfigProto
- func (*SavedModelConfig) ProtoMessage()
- func (x *SavedModelConfig) ProtoReflect() protoreflect.Message
- func (x *SavedModelConfig) Reset()
- func (x *SavedModelConfig) String() string
- type SessionBundleConfig
- func (*SessionBundleConfig) Descriptor() ([]byte, []int)deprecated
- func (x *SessionBundleConfig) GetBatchingParameters() *BatchingParameters
- func (x *SessionBundleConfig) GetEnableModelWarmup() bool
- func (x *SessionBundleConfig) GetEnablePerModelBatchingParams() bool
- func (x *SessionBundleConfig) GetEnableSavedModelConfig() bool
- func (x *SessionBundleConfig) GetEnableSessionMetadata() bool
- func (x *SessionBundleConfig) GetExperimentalFixedInputTensors() []*for_core_protos_go_proto.NamedTensorProto
- func (x *SessionBundleConfig) GetExperimentalTransientRamBytesDuringLoad() uint64
- func (x *SessionBundleConfig) GetMixedPrecision() string
- func (x *SessionBundleConfig) GetModelWarmupOptions() *ModelWarmupOptions
- func (x *SessionBundleConfig) GetNumTfliteInterpreters() int32deprecated
- func (x *SessionBundleConfig) GetNumTfliteInterpretersPerPool() int32
- func (x *SessionBundleConfig) GetNumTflitePools() int32
- func (x *SessionBundleConfig) GetPreferTfliteModel() bool
- func (x *SessionBundleConfig) GetRemoveUnusedFieldsFromBundleMetagraph() bool
- func (x *SessionBundleConfig) GetResourceEstimationUsesValidationResult() bool
- func (x *SessionBundleConfig) GetSavedModelTags() []string
- func (x *SessionBundleConfig) GetSessionConfig() *for_core_protos_go_proto.ConfigProto
- func (x *SessionBundleConfig) GetSessionRunLoadThreadpoolIndex() *wrapperspb.Int32Value
- func (x *SessionBundleConfig) GetSessionTarget() string
- func (x *SessionBundleConfig) GetWrapSessionWithNoThreadingParams() bool
- func (*SessionBundleConfig) ProtoMessage()
- func (x *SessionBundleConfig) ProtoReflect() protoreflect.Message
- func (x *SessionBundleConfig) Reset()
- func (x *SessionBundleConfig) String() string
- type SessionOverrides
- func (*SessionOverrides) Descriptor() ([]byte, []int)deprecated
- func (x *SessionOverrides) GetBatchOpRewriterConfig() *inference_go_proto.BatchOpRewriteConfig
- func (x *SessionOverrides) GetRemoteOpRemapConfig() *RemoteOpRemapConfig
- func (*SessionOverrides) ProtoMessage()
- func (x *SessionOverrides) ProtoReflect() protoreflect.Message
- func (x *SessionOverrides) Reset()
- func (x *SessionOverrides) String() string
- type TfrtSavedModelConfig
- func (*TfrtSavedModelConfig) Descriptor() ([]byte, []int)deprecated
- func (x *TfrtSavedModelConfig) GetAutoFusionMinClusterSize() int32deprecated
- func (x *TfrtSavedModelConfig) GetAutoFusionOp() []stringdeprecated
- func (x *TfrtSavedModelConfig) GetBatchingParameters() *BatchingParameters
- func (x *TfrtSavedModelConfig) GetEnableGrappler() bool
- func (x *TfrtSavedModelConfig) GetEnableGrapplerFunctionOptimizer() bool
- func (x *TfrtSavedModelConfig) GetEnableMlrt() bool
- func (x *TfrtSavedModelConfig) GetEnableModelWarmup() bool
- func (x *TfrtSavedModelConfig) GetEnableNativeOps() booldeprecated
- func (x *TfrtSavedModelConfig) GetEnableOnlineCostAnalysis() bool
- func (x *TfrtSavedModelConfig) GetEnablePaging() bool
- func (x *TfrtSavedModelConfig) GetEnablePerModelBatchingParams() bool
- func (x *TfrtSavedModelConfig) GetEnableSavedModelConfig() bool
- func (x *TfrtSavedModelConfig) GetEnableTfrtGpu() bool
- func (x *TfrtSavedModelConfig) GetEnableTfrtUseIfrtCoreSelector() bool
- func (x *TfrtSavedModelConfig) GetEnableWhileParallelIterations() bool
- func (x *TfrtSavedModelConfig) GetFreezeAfterInit() bool
- func (x *TfrtSavedModelConfig) GetGpuSystemMemorySizeInMb() int32
- func (x *TfrtSavedModelConfig) GetGraphOptions() *for_core_protos_go_proto.GraphOptions
- func (x *TfrtSavedModelConfig) GetHoistInvariantOps() bool
- func (x *TfrtSavedModelConfig) GetLazyInitThreshold() int32
- func (x *TfrtSavedModelConfig) GetLazyLoadingUseGraphExecutor() bool
- func (x *TfrtSavedModelConfig) GetLegacyConfig() *SessionBundleConfig
- func (x *TfrtSavedModelConfig) GetMaybeLoadFromMla() bool
- func (x *TfrtSavedModelConfig) GetMergeInterDependentStreams() bool
- func (x *TfrtSavedModelConfig) GetModelWarmupOptions() *ModelWarmupOptions
- func (x *TfrtSavedModelConfig) GetPredictResponseTensorSerializationOption() TfrtSavedModelConfig_PredictResponseTensorSerializationOption
- func (x *TfrtSavedModelConfig) GetResourceEstimationUsesValidationResult() bool
- func (x *TfrtSavedModelConfig) GetRunPlacerGrapplerOnFunctions() bool
- func (x *TfrtSavedModelConfig) GetSavedModelTags() []string
- func (x *TfrtSavedModelConfig) GetSinkInInvariantOps() bool
- func (x *TfrtSavedModelConfig) GetSkipWarmupRequestsIfInitialized() bool
- func (x *TfrtSavedModelConfig) GetStreamMergeThreshold() int64
- func (x *TfrtSavedModelConfig) GetStreamMergeUpperCostThreshold() int64
- func (x *TfrtSavedModelConfig) GetTargetTpu() bool
- func (x *TfrtSavedModelConfig) GetTfrtGpuParallelism() int32
- func (x *TfrtSavedModelConfig) GetTfrtMinNumBatchThreads() int64
- func (x *TfrtSavedModelConfig) GetTfrtUseFusedGpuOp() bool
- func (x *TfrtSavedModelConfig) GetTfrtUseIfrt() bool
- func (x *TfrtSavedModelConfig) GetThreadPoolFactoryConfigFilepath() string
- func (x *TfrtSavedModelConfig) GetTpuGatherTableWidthThresholdBytes() int32
- func (x *TfrtSavedModelConfig) GetTpuMoveResourceGatherToHost() bool
- func (x *TfrtSavedModelConfig) GetTpuUnpaddedBatchMode() TfrtSavedModelConfig_TpuUnpaddedBatchMode
- func (x *TfrtSavedModelConfig) GetUseFusedTpuOp() bool
- func (x *TfrtSavedModelConfig) GetUseLiteMlirImporter() bool
- func (x *TfrtSavedModelConfig) GetUseMira() bool
- func (x *TfrtSavedModelConfig) GetUseTpuHostAllocatorForInputs() bool
- func (x *TfrtSavedModelConfig) GetValidateInputSpecs() bool
- func (x *TfrtSavedModelConfig) GetValidateInputSpecsDryRun() bool
- func (*TfrtSavedModelConfig) ProtoMessage()
- func (x *TfrtSavedModelConfig) ProtoReflect() protoreflect.Message
- func (x *TfrtSavedModelConfig) Reset()
- func (x *TfrtSavedModelConfig) String() string
- type TfrtSavedModelConfig_PredictResponseTensorSerializationOption
- func (TfrtSavedModelConfig_PredictResponseTensorSerializationOption) Descriptor() protoreflect.EnumDescriptor
- func (x TfrtSavedModelConfig_PredictResponseTensorSerializationOption) Enum() *TfrtSavedModelConfig_PredictResponseTensorSerializationOption
- func (TfrtSavedModelConfig_PredictResponseTensorSerializationOption) EnumDescriptor() ([]byte, []int)deprecated
- func (x TfrtSavedModelConfig_PredictResponseTensorSerializationOption) Number() protoreflect.EnumNumber
- func (x TfrtSavedModelConfig_PredictResponseTensorSerializationOption) String() string
- func (TfrtSavedModelConfig_PredictResponseTensorSerializationOption) Type() protoreflect.EnumType
- type TfrtSavedModelConfig_TpuUnpaddedBatchMode
- func (TfrtSavedModelConfig_TpuUnpaddedBatchMode) Descriptor() protoreflect.EnumDescriptor
- func (x TfrtSavedModelConfig_TpuUnpaddedBatchMode) Enum() *TfrtSavedModelConfig_TpuUnpaddedBatchMode
- func (TfrtSavedModelConfig_TpuUnpaddedBatchMode) EnumDescriptor() ([]byte, []int)deprecated
- func (x TfrtSavedModelConfig_TpuUnpaddedBatchMode) Number() protoreflect.EnumNumber
- func (x TfrtSavedModelConfig_TpuUnpaddedBatchMode) String() string
- func (TfrtSavedModelConfig_TpuUnpaddedBatchMode) Type() protoreflect.EnumType
- type TfrtSavedModelSourceAdapterConfig
- func (*TfrtSavedModelSourceAdapterConfig) Descriptor() ([]byte, []int)deprecated
- func (x *TfrtSavedModelSourceAdapterConfig) GetSavedModelConfig() *TfrtSavedModelConfig
- func (*TfrtSavedModelSourceAdapterConfig) ProtoMessage()
- func (x *TfrtSavedModelSourceAdapterConfig) ProtoReflect() protoreflect.Message
- func (x *TfrtSavedModelSourceAdapterConfig) Reset()
- func (x *TfrtSavedModelSourceAdapterConfig) String() string
- type ThreadPoolFactoryConfig
- func (*ThreadPoolFactoryConfig) Descriptor() ([]byte, []int)deprecated
- func (x *ThreadPoolFactoryConfig) GetThreadPoolFactoryConfig() *anypb.Any
- func (*ThreadPoolFactoryConfig) ProtoMessage()
- func (x *ThreadPoolFactoryConfig) ProtoReflect() protoreflect.Message
- func (x *ThreadPoolFactoryConfig) Reset()
- func (x *ThreadPoolFactoryConfig) String() string
Constants ¶
This section is empty.
Variables ¶
var ( TfrtSavedModelConfig_TpuUnpaddedBatchMode_name = map[int32]string{ 0: "UNPADDED_BATCH_DISABLED", 1: "UNPADDED_BATCH_AUTO", 2: "UNPADDED_BATCH_ENFORCED", } TfrtSavedModelConfig_TpuUnpaddedBatchMode_value = map[string]int32{ "UNPADDED_BATCH_DISABLED": 0, "UNPADDED_BATCH_AUTO": 1, "UNPADDED_BATCH_ENFORCED": 2, } )
Enum value maps for TfrtSavedModelConfig_TpuUnpaddedBatchMode.
var ( TfrtSavedModelConfig_PredictResponseTensorSerializationOption_name = map[int32]string{ 0: "AS_PROTO_DEFAULT", 1: "AS_PROTO_CONTENT", } TfrtSavedModelConfig_PredictResponseTensorSerializationOption_value = map[string]int32{ "AS_PROTO_DEFAULT": 0, "AS_PROTO_FIELD": 0, "AS_PROTO_CONTENT": 1, } )
Enum value maps for TfrtSavedModelConfig_PredictResponseTensorSerializationOption.
var File_tensorflow_serving_servables_tensorflow_remote_op_config_rewriter_proto protoreflect.FileDescriptor
var File_tensorflow_serving_servables_tensorflow_saved_model_bundle_source_adapter_proto protoreflect.FileDescriptor
var File_tensorflow_serving_servables_tensorflow_saved_model_config_proto protoreflect.FileDescriptor
var File_tensorflow_serving_servables_tensorflow_session_bundle_config_proto protoreflect.FileDescriptor
var File_tensorflow_serving_servables_tensorflow_tfrt_saved_model_source_adapter_proto protoreflect.FileDescriptor
var File_tensorflow_serving_servables_tensorflow_thread_pool_factory_config_proto protoreflect.FileDescriptor
Functions ¶
This section is empty.
Types ¶
type BatchingParameters ¶
type BatchingParameters struct { // The maximum size of each input batch. // // IMPORTANT: As discussed above, use 'max_batch_size * 2' client threads to // achieve high throughput with batching. MaxBatchSize *wrapperspb.Int64Value `protobuf:"bytes,1,opt,name=max_batch_size,json=maxBatchSize,proto3" json:"max_batch_size,omitempty"` // If a task has been enqueued for this amount of time (in microseconds), and // a thread is available, the scheduler will immediately form a batch from // enqueued tasks and assign the batch to the thread for processing, even if // the batch's size is below 'max_batch_size'. BatchTimeoutMicros *wrapperspb.Int64Value `protobuf:"bytes,2,opt,name=batch_timeout_micros,json=batchTimeoutMicros,proto3" json:"batch_timeout_micros,omitempty"` // The maximum length of the queue, in terms of the number of batches. (A // batch that has been scheduled on a thread is considered to have been // removed from the queue.) MaxEnqueuedBatches *wrapperspb.Int64Value `protobuf:"bytes,3,opt,name=max_enqueued_batches,json=maxEnqueuedBatches,proto3" json:"max_enqueued_batches,omitempty"` // The number of threads to use to process batches. // Must be >= 1, and should be tuned carefully. NumBatchThreads *wrapperspb.Int64Value `protobuf:"bytes,4,opt,name=num_batch_threads,json=numBatchThreads,proto3" json:"num_batch_threads,omitempty"` // The name to use for the pool of batch threads. ThreadPoolName *wrapperspb.StringValue `protobuf:"bytes,5,opt,name=thread_pool_name,json=threadPoolName,proto3" json:"thread_pool_name,omitempty"` // If true, queue implementation would split one input batch task into // subtasks (as specified by `split_input_task_func` below) and fit subtasks // into different batches. EnableLargeBatchSplitting *wrapperspb.BoolValue `` /* 140-byte string literal not displayed */ // The maximum size of each enqueued batch to be processed (i.e., in // `batches_`). Relevant iff enable_large_batch_splitting is true. And when // relevant, 'max_batch_size' should be greater or equal than // `max_execution_batch_size` // // The scheduler may form batches of any size between 1 and this number // (inclusive). MaxExecutionBatchSize *wrapperspb.Int64Value `` /* 128-byte string literal not displayed */ // The allowed batch sizes. (Ignored if left empty.) // Requirements: // - The entries must be in increasing order. // - The final entry must equal 'max_batch_size'. AllowedBatchSizes []int64 `protobuf:"varint,6,rep,packed,name=allowed_batch_sizes,json=allowedBatchSizes,proto3" json:"allowed_batch_sizes,omitempty"` // Whether to pad variable-length inputs when a batch is formed. PadVariableLengthInputs bool `` /* 135-byte string literal not displayed */ // contains filtered or unexported fields }
Batching parameters. Each individual parameter is optional. If omitted, the default value from the relevant batching config struct (SharedBatchScheduler ::Options or BatchSchedulerRetrier::Options) is used.
func (*BatchingParameters) Descriptor
deprecated
func (*BatchingParameters) Descriptor() ([]byte, []int)
Deprecated: Use BatchingParameters.ProtoReflect.Descriptor instead.
func (*BatchingParameters) GetAllowedBatchSizes ¶
func (x *BatchingParameters) GetAllowedBatchSizes() []int64
func (*BatchingParameters) GetBatchTimeoutMicros ¶
func (x *BatchingParameters) GetBatchTimeoutMicros() *wrapperspb.Int64Value
func (*BatchingParameters) GetEnableLargeBatchSplitting ¶
func (x *BatchingParameters) GetEnableLargeBatchSplitting() *wrapperspb.BoolValue
func (*BatchingParameters) GetMaxBatchSize ¶
func (x *BatchingParameters) GetMaxBatchSize() *wrapperspb.Int64Value
func (*BatchingParameters) GetMaxEnqueuedBatches ¶
func (x *BatchingParameters) GetMaxEnqueuedBatches() *wrapperspb.Int64Value
func (*BatchingParameters) GetMaxExecutionBatchSize ¶
func (x *BatchingParameters) GetMaxExecutionBatchSize() *wrapperspb.Int64Value
func (*BatchingParameters) GetNumBatchThreads ¶
func (x *BatchingParameters) GetNumBatchThreads() *wrapperspb.Int64Value
func (*BatchingParameters) GetPadVariableLengthInputs ¶
func (x *BatchingParameters) GetPadVariableLengthInputs() bool
func (*BatchingParameters) GetThreadPoolName ¶
func (x *BatchingParameters) GetThreadPoolName() *wrapperspb.StringValue
func (*BatchingParameters) ProtoMessage ¶
func (*BatchingParameters) ProtoMessage()
func (*BatchingParameters) ProtoReflect ¶
func (x *BatchingParameters) ProtoReflect() protoreflect.Message
func (*BatchingParameters) Reset ¶
func (x *BatchingParameters) Reset()
func (*BatchingParameters) String ¶
func (x *BatchingParameters) String() string
type ModelWarmupOptions ¶
type ModelWarmupOptions struct { // Number of times a request is iterated during warmup replay. By default 1. NumRequestIterations *wrapperspb.Int32Value `protobuf:"bytes,1,opt,name=num_request_iterations,json=numRequestIterations,proto3" json:"num_request_iterations,omitempty"` // The number of threads to parallel execute warm up queries. By default 1. NumModelWarmupThreads *wrapperspb.Int32Value `` /* 128-byte string literal not displayed */ // Model name. ModelName string `protobuf:"bytes,3,opt,name=model_name,json=modelName,proto3" json:"model_name,omitempty"` // Model version. ModelVersion int64 `protobuf:"varint,4,opt,name=model_version,json=modelVersion,proto3" json:"model_version,omitempty"` // If true, warmup queries initiate parallel (dummy) warmup queries for each // `allowed_batch_sizes` of supported batch ops. // The extra queries' outputs are not returned. EnableAllBatchSizesWarmup bool `` /* 143-byte string literal not displayed */ // contains filtered or unexported fields }
Options related to model-warmup.
func (*ModelWarmupOptions) Descriptor
deprecated
func (*ModelWarmupOptions) Descriptor() ([]byte, []int)
Deprecated: Use ModelWarmupOptions.ProtoReflect.Descriptor instead.
func (*ModelWarmupOptions) GetEnableAllBatchSizesWarmup ¶
func (x *ModelWarmupOptions) GetEnableAllBatchSizesWarmup() bool
func (*ModelWarmupOptions) GetModelName ¶
func (x *ModelWarmupOptions) GetModelName() string
func (*ModelWarmupOptions) GetModelVersion ¶
func (x *ModelWarmupOptions) GetModelVersion() int64
func (*ModelWarmupOptions) GetNumModelWarmupThreads ¶
func (x *ModelWarmupOptions) GetNumModelWarmupThreads() *wrapperspb.Int32Value
func (*ModelWarmupOptions) GetNumRequestIterations ¶
func (x *ModelWarmupOptions) GetNumRequestIterations() *wrapperspb.Int32Value
func (*ModelWarmupOptions) ProtoMessage ¶
func (*ModelWarmupOptions) ProtoMessage()
func (*ModelWarmupOptions) ProtoReflect ¶
func (x *ModelWarmupOptions) ProtoReflect() protoreflect.Message
func (*ModelWarmupOptions) Reset ¶
func (x *ModelWarmupOptions) Reset()
func (*ModelWarmupOptions) String ¶
func (x *ModelWarmupOptions) String() string
type RemoteOpRemapConfig ¶
type RemoteOpRemapConfig struct { // Map for placeholder target model_names to actual target model_name. ModelNameRemap map[string]string `` /* 193-byte string literal not displayed */ // Map for placeholder target_address to actual target_address. TargetAddressRemap map[string]string `` /* 205-byte string literal not displayed */ // Map for model name to actual target_address. This will only be used when // target_address_remap is not set. Also, when the model_name_remap is set, // the model name here is the name BEFORE the rewrite. The model name here is // backend model name. BackendModelNameToTargetAddressRemap map[string]string `` /* 267-byte string literal not displayed */ // If true, version will be propagated from SessionMetadata.version. // See: http://shortn/_XDBisC2j9k // Requires enable_tf_session_metadata = true. PropagateVersion bool `protobuf:"varint,4,opt,name=propagate_version,json=propagateVersion,proto3" json:"propagate_version,omitempty"` // If true, model name will be propagated from SessionMetadata.name. // See: http://shortn/_XDBisC2j9k // Requires enable_tf_session_metadata = true. This takes precedence // over remapping from model_name_remap. PropagateModelName bool `protobuf:"varint,5,opt,name=propagate_model_name,json=propagateModelName,proto3" json:"propagate_model_name,omitempty"` // contains filtered or unexported fields }
Config for the remote op rewriter. This should be serialized/encoded and set a param in RewriterConfig with key kRemoteOpRewriteConfigParamKey.
func (*RemoteOpRemapConfig) Descriptor
deprecated
func (*RemoteOpRemapConfig) Descriptor() ([]byte, []int)
Deprecated: Use RemoteOpRemapConfig.ProtoReflect.Descriptor instead.
func (*RemoteOpRemapConfig) GetBackendModelNameToTargetAddressRemap ¶
func (x *RemoteOpRemapConfig) GetBackendModelNameToTargetAddressRemap() map[string]string
func (*RemoteOpRemapConfig) GetModelNameRemap ¶
func (x *RemoteOpRemapConfig) GetModelNameRemap() map[string]string
func (*RemoteOpRemapConfig) GetPropagateModelName ¶
func (x *RemoteOpRemapConfig) GetPropagateModelName() bool
func (*RemoteOpRemapConfig) GetPropagateVersion ¶
func (x *RemoteOpRemapConfig) GetPropagateVersion() bool
func (*RemoteOpRemapConfig) GetTargetAddressRemap ¶
func (x *RemoteOpRemapConfig) GetTargetAddressRemap() map[string]string
func (*RemoteOpRemapConfig) ProtoMessage ¶
func (*RemoteOpRemapConfig) ProtoMessage()
func (*RemoteOpRemapConfig) ProtoReflect ¶
func (x *RemoteOpRemapConfig) ProtoReflect() protoreflect.Message
func (*RemoteOpRemapConfig) Reset ¶
func (x *RemoteOpRemapConfig) Reset()
func (*RemoteOpRemapConfig) String ¶
func (x *RemoteOpRemapConfig) String() string
type SavedModelBundleSourceAdapterConfig ¶
type SavedModelBundleSourceAdapterConfig struct { // A SessionBundleConfig. // FOR INTERNAL USE ONLY DURING TRANSITION TO SAVED_MODEL. WILL BE DEPRECATED. // TODO(b/32248363): Replace this field with the "real" field(s). LegacyConfig *SessionBundleConfig `protobuf:"bytes,1000,opt,name=legacy_config,json=legacyConfig,proto3" json:"legacy_config,omitempty"` // contains filtered or unexported fields }
Config proto for SavedModelBundleSourceAdapter.
func (*SavedModelBundleSourceAdapterConfig) Descriptor
deprecated
func (*SavedModelBundleSourceAdapterConfig) Descriptor() ([]byte, []int)
Deprecated: Use SavedModelBundleSourceAdapterConfig.ProtoReflect.Descriptor instead.
func (*SavedModelBundleSourceAdapterConfig) GetLegacyConfig ¶
func (x *SavedModelBundleSourceAdapterConfig) GetLegacyConfig() *SessionBundleConfig
func (*SavedModelBundleSourceAdapterConfig) ProtoMessage ¶
func (*SavedModelBundleSourceAdapterConfig) ProtoMessage()
func (*SavedModelBundleSourceAdapterConfig) ProtoReflect ¶
func (x *SavedModelBundleSourceAdapterConfig) ProtoReflect() protoreflect.Message
func (*SavedModelBundleSourceAdapterConfig) Reset ¶
func (x *SavedModelBundleSourceAdapterConfig) Reset()
func (*SavedModelBundleSourceAdapterConfig) String ¶
func (x *SavedModelBundleSourceAdapterConfig) String() string
type SavedModelConfig ¶
type SavedModelConfig struct { // A select set of fields from SessionOptions which, at the model level, can // be used to override SessionOptions set for the entire processes. SessionOverrides *SessionOverrides `protobuf:"bytes,1,opt,name=session_overrides,json=sessionOverrides,proto3,oneof" json:"session_overrides,omitempty"` // Runtime specific configuration proto. They can be used to specify // environments for runtime. TfrtRuntimeConfig *graph_executor_go_proto.RuntimeConfigProto `protobuf:"bytes,2,opt,name=tfrt_runtime_config,json=tfrtRuntimeConfig,proto3,oneof" json:"tfrt_runtime_config,omitempty"` // A boolean field that indicates whether the model is critical, i.e., whether // the entire server cannot serve requests before this model has been loaded. Critical bool `protobuf:"varint,3,opt,name=critical,proto3" json:"critical,omitempty"` // contains filtered or unexported fields }
func (*SavedModelConfig) Descriptor
deprecated
func (*SavedModelConfig) Descriptor() ([]byte, []int)
Deprecated: Use SavedModelConfig.ProtoReflect.Descriptor instead.
func (*SavedModelConfig) GetCritical ¶
func (x *SavedModelConfig) GetCritical() bool
func (*SavedModelConfig) GetSessionOverrides ¶
func (x *SavedModelConfig) GetSessionOverrides() *SessionOverrides
func (*SavedModelConfig) GetTfrtRuntimeConfig ¶
func (x *SavedModelConfig) GetTfrtRuntimeConfig() *graph_executor_go_proto.RuntimeConfigProto
func (*SavedModelConfig) ProtoMessage ¶
func (*SavedModelConfig) ProtoMessage()
func (*SavedModelConfig) ProtoReflect ¶
func (x *SavedModelConfig) ProtoReflect() protoreflect.Message
func (*SavedModelConfig) Reset ¶
func (x *SavedModelConfig) Reset()
func (*SavedModelConfig) String ¶
func (x *SavedModelConfig) String() string
type SessionBundleConfig ¶
type SessionBundleConfig struct { // The TensorFlow runtime to connect to. // See full documentation in tensorflow/core/public/session_options.h. // // For single machine serving, we recommend using the empty string "", which // will configure the local TensorFlow runtime implementation. This provides // the best isolation currently available across multiple Session servables. SessionTarget string `protobuf:"bytes,1,opt,name=session_target,json=sessionTarget,proto3" json:"session_target,omitempty"` // TensorFlow Session configuration options. // See details at tensorflow/core/protobuf/config.proto. SessionConfig *for_core_protos_go_proto.ConfigProto `protobuf:"bytes,2,opt,name=session_config,json=sessionConfig,proto3" json:"session_config,omitempty"` // If set, each emitted session is wrapped with a layer that schedules Run() // calls in batches. The batching layer is transparent to the client // (implements the tensorflow::Session API). // // IMPORTANT: With batching enabled, client threads will spend most of their // time blocked on Session::Run() calls, waiting for enough peer threads to // also call Session::Run() such that a large batch can be formed. For good // throughput, we recommend setting the number of client threads equal to // roughly twice the maximum batch size ('max_batch_size' below). // // The batching layer uses a SharedBatchScheduler to coordinate batching // across multiple session servables emitted by this source adapter. A // BatchSchedulerRetrier is added on top of each batching session. BatchingParameters *BatchingParameters `protobuf:"bytes,3,opt,name=batching_parameters,json=batchingParameters,proto3" json:"batching_parameters,omitempty"` // If set, session run calls use a separate threadpool for restore and init // ops as part of loading the session-bundle. The value of this field should // correspond to the index of the tensorflow::ThreadPoolOptionProto defined as // part of `session_config.session_inter_op_thread_pool`. SessionRunLoadThreadpoolIndex *wrapperspb.Int32Value `` /* 154-byte string literal not displayed */ // EXPERIMENTAL. THIS FIELD MAY CHANGE OR GO AWAY. USE WITH CAUTION. // // Transient memory used while loading a model, which is released once the // loading phase has completed. (This is on top of the memory used in steady- // state while the model is in memory after it has finished loading.) // // TODO(b/38376838): This is a temporary hack, and it applies to all models. // Remove it once resource estimates are moved inside SavedModel. ExperimentalTransientRamBytesDuringLoad uint64 `` /* 187-byte string literal not displayed */ // Set of SavedModel tags identifying the specific meta graph def to be // loaded. SavedModelTags []string `protobuf:"bytes,6,rep,name=saved_model_tags,json=savedModelTags,proto3" json:"saved_model_tags,omitempty"` // EXPERIMENTAL. THIS FIELD MAY CHANGE OR GO AWAY. USE WITH CAUTION. // // Input tensors to append to every Session::Run() call. ExperimentalFixedInputTensors []*for_core_protos_go_proto.NamedTensorProto `` /* 154-byte string literal not displayed */ // Enables model warmup. EnableModelWarmup bool `protobuf:"varint,779,opt,name=enable_model_warmup,json=enableModelWarmup,proto3" json:"enable_model_warmup,omitempty"` ModelWarmupOptions *ModelWarmupOptions `protobuf:"bytes,780,opt,name=model_warmup_options,json=modelWarmupOptions,proto3" json:"model_warmup_options,omitempty"` // EXPERIMENTAL. THIS FIELD MAY CHANGE OR GO AWAY. USE WITH CAUTION. // // Enables passing in the SessionMetadata to the Session. The SessionMetadata // consists of information like the model name, version, which can then be // used by the TensorFlow runtime appropriately (for debugging, logging, etc). EnableSessionMetadata bool `` /* 129-byte string literal not displayed */ // EXPERIMENTAL. THIS FIELD MAY CHANGE OR GO AWAY. USE WITH CAUTION. // // Remove unused fields from metagraph proto message in SavedModelBundle. // This message is stored alongside the `Session` object. Removing unwanted // fields helps reduce memory footprint. RemoveUnusedFieldsFromBundleMetagraph bool `` /* 183-byte string literal not displayed */ // EXPERIMENTAL. THIS FIELD MAY CHANGE OR GO AWAY. USE WITH CAUTION. // // Prefer TensorFlow Lite model from `model.tflite` file in SavedModel // directory, instead of the TensorFlow model from `saved_model.pb` file. // If no TensorFlow Lite model found, fallback to TensorFlow model. PreferTfliteModel bool `protobuf:"varint,783,opt,name=prefer_tflite_model,json=preferTfliteModel,proto3" json:"prefer_tflite_model,omitempty"` // Tries to use infra validation result to estimate resource usage. ResourceEstimationUsesValidationResult bool `` /* 184-byte string literal not displayed */ // Deprecated: Marked as deprecated in tensorflow/serving/servables/tensorflow/session_bundle_config.proto. NumTfliteInterpreters int32 `` /* 129-byte string literal not displayed */ // EXPERIMENTAL. THIS FIELD MAY CHANGE OR GO AWAY. USE WITH CAUTION. // // Number of TFLite interpreters in an interpreter pool of TfLiteSession. NumTfliteInterpretersPerPool int32 `` /* 154-byte string literal not displayed */ // EXPERIMENTAL. THIS FIELD MAY CHANGE OR GO AWAY. USE WITH CAUTION. // // Number of TFLite interpreter pools in a TfLiteSession. NumTflitePools int32 `protobuf:"varint,787,opt,name=num_tflite_pools,json=numTflitePools,proto3" json:"num_tflite_pools,omitempty"` // EXPERIMENTAL. THIS FIELD MAY CHANGE OR GO AWAY. USE WITH CAUTION. // // Use SessionWrapperIgnoreThreadPoolOptions instead. WrapSessionWithNoThreadingParams bool `` /* 168-byte string literal not displayed */ // EXPERIMENTAL. THIS FIELD MAY CHANGE OR GO AWAY. USE WITH CAUTION. // // Enable per-model batching parameters (present in SavedModel). If this // option is enabled, model specific batching params (e.g. timeout, batch // sizes etc.) from `batching_parameters` field above are *ignored* and // instead the one in SavedModel directory are used. This field is only // used if batching is enabled (i.e. `batching_parameters` message above // is set). EnablePerModelBatchingParams bool `` /* 154-byte string literal not displayed */ // EXPERIMENTAL. THIS FIELD MAY CHANGE OR GO AWAY. USE WITH CAUTION // // Enable overriding model configs via assets.extra/saved_model_config.pb. EnableSavedModelConfig bool `` /* 134-byte string literal not displayed */ // Add bf16 mixed_precision option MixedPrecision string `protobuf:"bytes,791,opt,name=mixed_precision,json=mixedPrecision,proto3" json:"mixed_precision,omitempty"` // contains filtered or unexported fields }
Configuration parameters for a SessionBundle, with optional batching.
func (*SessionBundleConfig) Descriptor
deprecated
func (*SessionBundleConfig) Descriptor() ([]byte, []int)
Deprecated: Use SessionBundleConfig.ProtoReflect.Descriptor instead.
func (*SessionBundleConfig) GetBatchingParameters ¶
func (x *SessionBundleConfig) GetBatchingParameters() *BatchingParameters
func (*SessionBundleConfig) GetEnableModelWarmup ¶
func (x *SessionBundleConfig) GetEnableModelWarmup() bool
func (*SessionBundleConfig) GetEnablePerModelBatchingParams ¶
func (x *SessionBundleConfig) GetEnablePerModelBatchingParams() bool
func (*SessionBundleConfig) GetEnableSavedModelConfig ¶
func (x *SessionBundleConfig) GetEnableSavedModelConfig() bool
func (*SessionBundleConfig) GetEnableSessionMetadata ¶
func (x *SessionBundleConfig) GetEnableSessionMetadata() bool
func (*SessionBundleConfig) GetExperimentalFixedInputTensors ¶
func (x *SessionBundleConfig) GetExperimentalFixedInputTensors() []*for_core_protos_go_proto.NamedTensorProto
func (*SessionBundleConfig) GetExperimentalTransientRamBytesDuringLoad ¶
func (x *SessionBundleConfig) GetExperimentalTransientRamBytesDuringLoad() uint64
func (*SessionBundleConfig) GetMixedPrecision ¶
func (x *SessionBundleConfig) GetMixedPrecision() string
func (*SessionBundleConfig) GetModelWarmupOptions ¶
func (x *SessionBundleConfig) GetModelWarmupOptions() *ModelWarmupOptions
func (*SessionBundleConfig) GetNumTfliteInterpreters
deprecated
func (x *SessionBundleConfig) GetNumTfliteInterpreters() int32
Deprecated: Marked as deprecated in tensorflow/serving/servables/tensorflow/session_bundle_config.proto.
func (*SessionBundleConfig) GetNumTfliteInterpretersPerPool ¶
func (x *SessionBundleConfig) GetNumTfliteInterpretersPerPool() int32
func (*SessionBundleConfig) GetNumTflitePools ¶
func (x *SessionBundleConfig) GetNumTflitePools() int32
func (*SessionBundleConfig) GetPreferTfliteModel ¶
func (x *SessionBundleConfig) GetPreferTfliteModel() bool
func (*SessionBundleConfig) GetRemoveUnusedFieldsFromBundleMetagraph ¶
func (x *SessionBundleConfig) GetRemoveUnusedFieldsFromBundleMetagraph() bool
func (*SessionBundleConfig) GetResourceEstimationUsesValidationResult ¶
func (x *SessionBundleConfig) GetResourceEstimationUsesValidationResult() bool
func (*SessionBundleConfig) GetSavedModelTags ¶
func (x *SessionBundleConfig) GetSavedModelTags() []string
func (*SessionBundleConfig) GetSessionConfig ¶
func (x *SessionBundleConfig) GetSessionConfig() *for_core_protos_go_proto.ConfigProto
func (*SessionBundleConfig) GetSessionRunLoadThreadpoolIndex ¶
func (x *SessionBundleConfig) GetSessionRunLoadThreadpoolIndex() *wrapperspb.Int32Value
func (*SessionBundleConfig) GetSessionTarget ¶
func (x *SessionBundleConfig) GetSessionTarget() string
func (*SessionBundleConfig) GetWrapSessionWithNoThreadingParams ¶
func (x *SessionBundleConfig) GetWrapSessionWithNoThreadingParams() bool
func (*SessionBundleConfig) ProtoMessage ¶
func (*SessionBundleConfig) ProtoMessage()
func (*SessionBundleConfig) ProtoReflect ¶
func (x *SessionBundleConfig) ProtoReflect() protoreflect.Message
func (*SessionBundleConfig) Reset ¶
func (x *SessionBundleConfig) Reset()
func (*SessionBundleConfig) String ¶
func (x *SessionBundleConfig) String() string
type SessionOverrides ¶
type SessionOverrides struct { RemoteOpRemapConfig *RemoteOpRemapConfig `` /* 128-byte string literal not displayed */ BatchOpRewriterConfig *inference_go_proto.BatchOpRewriteConfig `` /* 134-byte string literal not displayed */ // contains filtered or unexported fields }
func (*SessionOverrides) Descriptor
deprecated
func (*SessionOverrides) Descriptor() ([]byte, []int)
Deprecated: Use SessionOverrides.ProtoReflect.Descriptor instead.
func (*SessionOverrides) GetBatchOpRewriterConfig ¶
func (x *SessionOverrides) GetBatchOpRewriterConfig() *inference_go_proto.BatchOpRewriteConfig
func (*SessionOverrides) GetRemoteOpRemapConfig ¶
func (x *SessionOverrides) GetRemoteOpRemapConfig() *RemoteOpRemapConfig
func (*SessionOverrides) ProtoMessage ¶
func (*SessionOverrides) ProtoMessage()
func (*SessionOverrides) ProtoReflect ¶
func (x *SessionOverrides) ProtoReflect() protoreflect.Message
func (*SessionOverrides) Reset ¶
func (x *SessionOverrides) Reset()
func (*SessionOverrides) String ¶
func (x *SessionOverrides) String() string
type TfrtSavedModelConfig ¶
type TfrtSavedModelConfig struct { // A SessionBundleConfig. // Deprecated. LegacyConfig *SessionBundleConfig `protobuf:"bytes,1000,opt,name=legacy_config,json=legacyConfig,proto3" json:"legacy_config,omitempty"` // If true, native ops will be used if they are implemented in TFRT. If // false, all ops are using fallback. // // Deprecated: Marked as deprecated in tensorflow/serving/servables/tensorflow/tfrt_saved_model_source_adapter.proto. EnableNativeOps bool `protobuf:"varint,1001,opt,name=enable_native_ops,json=enableNativeOps,proto3" json:"enable_native_ops,omitempty"` // If true, the lite version of MLIR importer would be used. UseLiteMlirImporter bool `` /* 126-byte string literal not displayed */ // If true, enable grappler for TFRT. EnableGrappler bool `protobuf:"varint,1003,opt,name=enable_grappler,json=enableGrappler,proto3" json:"enable_grappler,omitempty"` // If true, when creating an optimized subgraph, Placer and Grappler will // also run on the functions. RunPlacerGrapplerOnFunctions bool `` /* 155-byte string literal not displayed */ // If true, the function optimizer in the grappler will be enabled, and // optimizations like function inlining will be applied. EnableGrapplerFunctionOptimizer bool `` /* 162-byte string literal not displayed */ // If true, for each client graph, the op costs of the first request will be // recorded and used to re-compile the client graph. EnableOnlineCostAnalysis bool `` /* 141-byte string literal not displayed */ // If true, the TFRT runtime will use the new interpreter for host execution. EnableMlrt bool `protobuf:"varint,1030,opt,name=enable_mlrt,json=enableMlrt,proto3" json:"enable_mlrt,omitempty"` // If true, TFRT will perform GPU related compiler passes. EnableTfrtGpu bool `protobuf:"varint,1024,opt,name=enable_tfrt_gpu,json=enableTfrtGpu,proto3" json:"enable_tfrt_gpu,omitempty"` // If true, TFRT will use TPU specific compiler passes and perform TPU // specific initialization. TargetTpu bool `protobuf:"varint,1004,opt,name=target_tpu,json=targetTpu,proto3" json:"target_tpu,omitempty"` // If true, resource loading will happen during model loading instead of // inference time. HoistInvariantOps bool `protobuf:"varint,1005,opt,name=hoist_invariant_ops,json=hoistInvariantOps,proto3" json:"hoist_invariant_ops,omitempty"` // If true, sink the resource handle into the nested function to facilitate // invariant ops hoisting. SinkInInvariantOps bool `protobuf:"varint,1028,opt,name=sink_in_invariant_ops,json=sinkInInvariantOps,proto3" json:"sink_in_invariant_ops,omitempty"` // The threshold to merge small streams in TFRT. The stream with cost // smaller than the threshold will be merged. StreamMergeThreshold int64 `` /* 127-byte string literal not displayed */ // The upper bond to merge dependent streams. StreamMergeUpperCostThreshold int64 `` /* 158-byte string literal not displayed */ // If true, streams with inter data depenedencies will be merged. MergeInterDependentStreams bool `` /* 147-byte string literal not displayed */ // Options that apply to all graphs. GraphOptions *for_core_protos_go_proto.GraphOptions `protobuf:"bytes,1006,opt,name=graph_options,json=graphOptions,proto3" json:"graph_options,omitempty"` // If set, each emitted saved model is wrapped with a layer that schedules // Run() calls in batches. The batching layer is transparent to the client // (implements the tfrt::SavedModel API). // // IMPORTANT: With batching enabled, client threads will spend most of their // time blocked on SavedModel::Run() calls, waiting for enough peer threads to // also call SavedModel::Run() such that a large batch can be formed. For good // throughput, we recommend setting the number of client threads equal to // roughly twice the maximum batch size ('max_batch_size' below). // // The batching layer uses a SharedBatchScheduler to coordinate batching // across multiple saved model servables emitted by this source adapter. A // BatchSchedulerRetrier is added on top of each batching saved model. BatchingParameters *BatchingParameters `protobuf:"bytes,1007,opt,name=batching_parameters,json=batchingParameters,proto3" json:"batching_parameters,omitempty"` // Set of SavedModel tags identifying the specific meta graph def to be // loaded. SavedModelTags []string `protobuf:"bytes,1008,rep,name=saved_model_tags,json=savedModelTags,proto3" json:"saved_model_tags,omitempty"` // Enables model warmup. EnableModelWarmup bool `protobuf:"varint,1009,opt,name=enable_model_warmup,json=enableModelWarmup,proto3" json:"enable_model_warmup,omitempty"` ModelWarmupOptions *ModelWarmupOptions `protobuf:"bytes,1010,opt,name=model_warmup_options,json=modelWarmupOptions,proto3" json:"model_warmup_options,omitempty"` // Tries to use infra validation result to estimate resource usage. ResourceEstimationUsesValidationResult bool `` /* 185-byte string literal not displayed */ // If the model contains more than lazy_init_threshold signature defs, use // lazy initilizaion as it is likely not all sigantures are used for serving. LazyInitThreshold int32 `protobuf:"varint,1013,opt,name=lazy_init_threshold,json=lazyInitThreshold,proto3" json:"lazy_init_threshold,omitempty"` // If true, we'll attempt to find MLArchive within the model source path. If // no MLArchive is found, will use the path as a normal SavedModel directory. MaybeLoadFromMla bool `protobuf:"varint,1025,opt,name=maybe_load_from_mla,json=maybeLoadFromMla,proto3" json:"maybe_load_from_mla,omitempty"` // If true, MIRA implementation of SavedModel API is used, so MIRA API is used // for loading and running the model. UseMira bool `protobuf:"varint,1027,opt,name=use_mira,json=useMira,proto3" json:"use_mira,omitempty"` // If true, skip the warmup requests if the signature def is initialized. SkipWarmupRequestsIfInitialized bool `` /* 164-byte string literal not displayed */ // If true, move resource gather op to host side. TpuMoveResourceGatherToHost bool `` /* 154-byte string literal not displayed */ // If tpu_move_resource_gather_to_host is true and the width of the gather // table is less than the threshold, it will be placed on the host. TpuGatherTableWidthThresholdBytes int32 `` /* 172-byte string literal not displayed */ // If true, use fuesd op for tpu compile, execute and data transfer. UseFusedTpuOp bool `protobuf:"varint,1018,opt,name=use_fused_tpu_op,json=useFusedTpuOp,proto3" json:"use_fused_tpu_op,omitempty"` // If true, tf.While's iterations will be parallelized. EnableWhileParallelIterations bool `` /* 156-byte string literal not displayed */ // If true, fallback executeops that produce inputs to tpu program will use // tpu host allocator. UseTpuHostAllocatorForInputs bool `` /* 157-byte string literal not displayed */ // Enables model specific runtime params described in saved model's // assets.extra/saved_model_config.pb file. EnableSavedModelConfig bool `` /* 135-byte string literal not displayed */ // These are no longer used. // // Deprecated: Marked as deprecated in tensorflow/serving/servables/tensorflow/tfrt_saved_model_source_adapter.proto. AutoFusionOp []string `protobuf:"bytes,2006,rep,name=auto_fusion_op,json=autoFusionOp,proto3" json:"auto_fusion_op,omitempty"` // Deprecated: Marked as deprecated in tensorflow/serving/servables/tensorflow/tfrt_saved_model_source_adapter.proto. AutoFusionMinClusterSize int32 `` /* 143-byte string literal not displayed */ // EXPERIMENTAL. THIS FIELD MAY CHANGE OR GO AWAY. USE WITH CAUTION. // // Enable per-model batching parameters (present in SavedModel). If this // option is enabled, model specific batching params (e.g. timeout, batch // sizes etc.) from `batching_parameters` field above are *ignored* and // instead the one in SavedModel directory are used. This field is only // used if batching is enabled (i.e. `batching_parameters` message above // is set). EnablePerModelBatchingParams bool `` /* 155-byte string literal not displayed */ // If true, the lazy loading path will use tfrt_stub::GraphExecutor. LazyLoadingUseGraphExecutor bool `` /* 152-byte string literal not displayed */ // Mode for TPU Unpadded batch. TpuUnpaddedBatchMode TfrtSavedModelConfig_TpuUnpaddedBatchMode `` /* 195-byte string literal not displayed */ // The config file path of the ThreadPoolFactory used to create inter-op // ThreadPools. If unspecified, the default Tensorflow threadpools will be // used. ThreadPoolFactoryConfigFilepath string `` /* 163-byte string literal not displayed */ // If true, validate input tensor specs before executing the request. ValidateInputSpecs bool `protobuf:"varint,2012,opt,name=validate_input_specs,json=validateInputSpecs,proto3" json:"validate_input_specs,omitempty"` // TODO(b/279197040) Remove after b/279197040 is fixed. // If true, validate (by logging mismatches) input tensor specs before // executing the request. ValidateInputSpecsDryRun bool `` /* 143-byte string literal not displayed */ PredictResponseTensorSerializationOption TfrtSavedModelConfig_PredictResponseTensorSerializationOption `` /* 277-byte string literal not displayed */ // The size of reserved memory space for GPU system. GpuSystemMemorySizeInMb int32 `` /* 142-byte string literal not displayed */ // Disables compilations after model initialization is complete // (ignored if enable_model_warmup is false) FreezeAfterInit bool `protobuf:"varint,2016,opt,name=freeze_after_init,json=freezeAfterInit,proto3" json:"freeze_after_init,omitempty"` // The number of virtual GPUs to create on a physical GPU. TfrtGpuParallelism int32 `protobuf:"varint,2017,opt,name=tfrt_gpu_parallelism,json=tfrtGpuParallelism,proto3" json:"tfrt_gpu_parallelism,omitempty"` // Whether to use fused op for GPU compile, execute and data transfer. TfrtUseFusedGpuOp bool `protobuf:"varint,2018,opt,name=tfrt_use_fused_gpu_op,json=tfrtUseFusedGpuOp,proto3" json:"tfrt_use_fused_gpu_op,omitempty"` // The minimum number of batch threads. TfrtMinNumBatchThreads int64 `` /* 137-byte string literal not displayed */ // Whether to use IFRT as device driver for device execution. TfrtUseIfrt bool `protobuf:"varint,2020,opt,name=tfrt_use_ifrt,json=tfrtUseIfrt,proto3" json:"tfrt_use_ifrt,omitempty"` // Whether to enable core selector for TFRT+IFRT. It is only effective when // tfrt_use_ifrt is true. EnableTfrtUseIfrtCoreSelector bool `` /* 160-byte string literal not displayed */ // Whether to enable paging. This should only be true when using Pathways // backend. EnablePaging bool `protobuf:"varint,2022,opt,name=enable_paging,json=enablePaging,proto3" json:"enable_paging,omitempty"` // contains filtered or unexported fields }
func (*TfrtSavedModelConfig) Descriptor
deprecated
func (*TfrtSavedModelConfig) Descriptor() ([]byte, []int)
Deprecated: Use TfrtSavedModelConfig.ProtoReflect.Descriptor instead.
func (*TfrtSavedModelConfig) GetAutoFusionMinClusterSize
deprecated
func (x *TfrtSavedModelConfig) GetAutoFusionMinClusterSize() int32
Deprecated: Marked as deprecated in tensorflow/serving/servables/tensorflow/tfrt_saved_model_source_adapter.proto.
func (*TfrtSavedModelConfig) GetAutoFusionOp
deprecated
func (x *TfrtSavedModelConfig) GetAutoFusionOp() []string
Deprecated: Marked as deprecated in tensorflow/serving/servables/tensorflow/tfrt_saved_model_source_adapter.proto.
func (*TfrtSavedModelConfig) GetBatchingParameters ¶
func (x *TfrtSavedModelConfig) GetBatchingParameters() *BatchingParameters
func (*TfrtSavedModelConfig) GetEnableGrappler ¶
func (x *TfrtSavedModelConfig) GetEnableGrappler() bool
func (*TfrtSavedModelConfig) GetEnableGrapplerFunctionOptimizer ¶
func (x *TfrtSavedModelConfig) GetEnableGrapplerFunctionOptimizer() bool
func (*TfrtSavedModelConfig) GetEnableMlrt ¶
func (x *TfrtSavedModelConfig) GetEnableMlrt() bool
func (*TfrtSavedModelConfig) GetEnableModelWarmup ¶
func (x *TfrtSavedModelConfig) GetEnableModelWarmup() bool
func (*TfrtSavedModelConfig) GetEnableNativeOps
deprecated
func (x *TfrtSavedModelConfig) GetEnableNativeOps() bool
Deprecated: Marked as deprecated in tensorflow/serving/servables/tensorflow/tfrt_saved_model_source_adapter.proto.
func (*TfrtSavedModelConfig) GetEnableOnlineCostAnalysis ¶
func (x *TfrtSavedModelConfig) GetEnableOnlineCostAnalysis() bool
func (*TfrtSavedModelConfig) GetEnablePaging ¶
func (x *TfrtSavedModelConfig) GetEnablePaging() bool
func (*TfrtSavedModelConfig) GetEnablePerModelBatchingParams ¶
func (x *TfrtSavedModelConfig) GetEnablePerModelBatchingParams() bool
func (*TfrtSavedModelConfig) GetEnableSavedModelConfig ¶
func (x *TfrtSavedModelConfig) GetEnableSavedModelConfig() bool
func (*TfrtSavedModelConfig) GetEnableTfrtGpu ¶
func (x *TfrtSavedModelConfig) GetEnableTfrtGpu() bool
func (*TfrtSavedModelConfig) GetEnableTfrtUseIfrtCoreSelector ¶
func (x *TfrtSavedModelConfig) GetEnableTfrtUseIfrtCoreSelector() bool
func (*TfrtSavedModelConfig) GetEnableWhileParallelIterations ¶
func (x *TfrtSavedModelConfig) GetEnableWhileParallelIterations() bool
func (*TfrtSavedModelConfig) GetFreezeAfterInit ¶
func (x *TfrtSavedModelConfig) GetFreezeAfterInit() bool
func (*TfrtSavedModelConfig) GetGpuSystemMemorySizeInMb ¶
func (x *TfrtSavedModelConfig) GetGpuSystemMemorySizeInMb() int32
func (*TfrtSavedModelConfig) GetGraphOptions ¶
func (x *TfrtSavedModelConfig) GetGraphOptions() *for_core_protos_go_proto.GraphOptions
func (*TfrtSavedModelConfig) GetHoistInvariantOps ¶
func (x *TfrtSavedModelConfig) GetHoistInvariantOps() bool
func (*TfrtSavedModelConfig) GetLazyInitThreshold ¶
func (x *TfrtSavedModelConfig) GetLazyInitThreshold() int32
func (*TfrtSavedModelConfig) GetLazyLoadingUseGraphExecutor ¶
func (x *TfrtSavedModelConfig) GetLazyLoadingUseGraphExecutor() bool
func (*TfrtSavedModelConfig) GetLegacyConfig ¶
func (x *TfrtSavedModelConfig) GetLegacyConfig() *SessionBundleConfig
func (*TfrtSavedModelConfig) GetMaybeLoadFromMla ¶
func (x *TfrtSavedModelConfig) GetMaybeLoadFromMla() bool
func (*TfrtSavedModelConfig) GetMergeInterDependentStreams ¶
func (x *TfrtSavedModelConfig) GetMergeInterDependentStreams() bool
func (*TfrtSavedModelConfig) GetModelWarmupOptions ¶
func (x *TfrtSavedModelConfig) GetModelWarmupOptions() *ModelWarmupOptions
func (*TfrtSavedModelConfig) GetPredictResponseTensorSerializationOption ¶
func (x *TfrtSavedModelConfig) GetPredictResponseTensorSerializationOption() TfrtSavedModelConfig_PredictResponseTensorSerializationOption
func (*TfrtSavedModelConfig) GetResourceEstimationUsesValidationResult ¶
func (x *TfrtSavedModelConfig) GetResourceEstimationUsesValidationResult() bool
func (*TfrtSavedModelConfig) GetRunPlacerGrapplerOnFunctions ¶
func (x *TfrtSavedModelConfig) GetRunPlacerGrapplerOnFunctions() bool
func (*TfrtSavedModelConfig) GetSavedModelTags ¶
func (x *TfrtSavedModelConfig) GetSavedModelTags() []string
func (*TfrtSavedModelConfig) GetSinkInInvariantOps ¶
func (x *TfrtSavedModelConfig) GetSinkInInvariantOps() bool
func (*TfrtSavedModelConfig) GetSkipWarmupRequestsIfInitialized ¶
func (x *TfrtSavedModelConfig) GetSkipWarmupRequestsIfInitialized() bool
func (*TfrtSavedModelConfig) GetStreamMergeThreshold ¶
func (x *TfrtSavedModelConfig) GetStreamMergeThreshold() int64
func (*TfrtSavedModelConfig) GetStreamMergeUpperCostThreshold ¶
func (x *TfrtSavedModelConfig) GetStreamMergeUpperCostThreshold() int64
func (*TfrtSavedModelConfig) GetTargetTpu ¶
func (x *TfrtSavedModelConfig) GetTargetTpu() bool
func (*TfrtSavedModelConfig) GetTfrtGpuParallelism ¶
func (x *TfrtSavedModelConfig) GetTfrtGpuParallelism() int32
func (*TfrtSavedModelConfig) GetTfrtMinNumBatchThreads ¶
func (x *TfrtSavedModelConfig) GetTfrtMinNumBatchThreads() int64
func (*TfrtSavedModelConfig) GetTfrtUseFusedGpuOp ¶
func (x *TfrtSavedModelConfig) GetTfrtUseFusedGpuOp() bool
func (*TfrtSavedModelConfig) GetTfrtUseIfrt ¶
func (x *TfrtSavedModelConfig) GetTfrtUseIfrt() bool
func (*TfrtSavedModelConfig) GetThreadPoolFactoryConfigFilepath ¶
func (x *TfrtSavedModelConfig) GetThreadPoolFactoryConfigFilepath() string
func (*TfrtSavedModelConfig) GetTpuGatherTableWidthThresholdBytes ¶
func (x *TfrtSavedModelConfig) GetTpuGatherTableWidthThresholdBytes() int32
func (*TfrtSavedModelConfig) GetTpuMoveResourceGatherToHost ¶
func (x *TfrtSavedModelConfig) GetTpuMoveResourceGatherToHost() bool
func (*TfrtSavedModelConfig) GetTpuUnpaddedBatchMode ¶
func (x *TfrtSavedModelConfig) GetTpuUnpaddedBatchMode() TfrtSavedModelConfig_TpuUnpaddedBatchMode
func (*TfrtSavedModelConfig) GetUseFusedTpuOp ¶
func (x *TfrtSavedModelConfig) GetUseFusedTpuOp() bool
func (*TfrtSavedModelConfig) GetUseLiteMlirImporter ¶
func (x *TfrtSavedModelConfig) GetUseLiteMlirImporter() bool
func (*TfrtSavedModelConfig) GetUseMira ¶
func (x *TfrtSavedModelConfig) GetUseMira() bool
func (*TfrtSavedModelConfig) GetUseTpuHostAllocatorForInputs ¶
func (x *TfrtSavedModelConfig) GetUseTpuHostAllocatorForInputs() bool
func (*TfrtSavedModelConfig) GetValidateInputSpecs ¶
func (x *TfrtSavedModelConfig) GetValidateInputSpecs() bool
func (*TfrtSavedModelConfig) GetValidateInputSpecsDryRun ¶
func (x *TfrtSavedModelConfig) GetValidateInputSpecsDryRun() bool
func (*TfrtSavedModelConfig) ProtoMessage ¶
func (*TfrtSavedModelConfig) ProtoMessage()
func (*TfrtSavedModelConfig) ProtoReflect ¶
func (x *TfrtSavedModelConfig) ProtoReflect() protoreflect.Message
func (*TfrtSavedModelConfig) Reset ¶
func (x *TfrtSavedModelConfig) Reset()
func (*TfrtSavedModelConfig) String ¶
func (x *TfrtSavedModelConfig) String() string
type TfrtSavedModelConfig_PredictResponseTensorSerializationOption ¶
type TfrtSavedModelConfig_PredictResponseTensorSerializationOption int32
In a predict handler, this option specifies how to serialize tensors (e.g: as proto fields or as proto content). Serialize as proto fields by default, for backward compatibility.
const ( TfrtSavedModelConfig_AS_PROTO_DEFAULT TfrtSavedModelConfig_PredictResponseTensorSerializationOption = 0 TfrtSavedModelConfig_AS_PROTO_FIELD TfrtSavedModelConfig_PredictResponseTensorSerializationOption = 0 TfrtSavedModelConfig_AS_PROTO_CONTENT TfrtSavedModelConfig_PredictResponseTensorSerializationOption = 1 )
func (TfrtSavedModelConfig_PredictResponseTensorSerializationOption) Descriptor ¶
func (TfrtSavedModelConfig_PredictResponseTensorSerializationOption) Descriptor() protoreflect.EnumDescriptor
func (TfrtSavedModelConfig_PredictResponseTensorSerializationOption) EnumDescriptor
deprecated
func (TfrtSavedModelConfig_PredictResponseTensorSerializationOption) EnumDescriptor() ([]byte, []int)
Deprecated: Use TfrtSavedModelConfig_PredictResponseTensorSerializationOption.Descriptor instead.
func (TfrtSavedModelConfig_PredictResponseTensorSerializationOption) String ¶
func (x TfrtSavedModelConfig_PredictResponseTensorSerializationOption) String() string
type TfrtSavedModelConfig_TpuUnpaddedBatchMode ¶
type TfrtSavedModelConfig_TpuUnpaddedBatchMode int32
const ( // Disable this feature. TfrtSavedModelConfig_UNPADDED_BATCH_DISABLED TfrtSavedModelConfig_TpuUnpaddedBatchMode = 0 // Enable this feature when in-graph batching is detected. TfrtSavedModelConfig_UNPADDED_BATCH_AUTO TfrtSavedModelConfig_TpuUnpaddedBatchMode = 1 // Always enable this feature. TfrtSavedModelConfig_UNPADDED_BATCH_ENFORCED TfrtSavedModelConfig_TpuUnpaddedBatchMode = 2 )
func (TfrtSavedModelConfig_TpuUnpaddedBatchMode) Descriptor ¶
func (TfrtSavedModelConfig_TpuUnpaddedBatchMode) Descriptor() protoreflect.EnumDescriptor
func (TfrtSavedModelConfig_TpuUnpaddedBatchMode) EnumDescriptor
deprecated
func (TfrtSavedModelConfig_TpuUnpaddedBatchMode) EnumDescriptor() ([]byte, []int)
Deprecated: Use TfrtSavedModelConfig_TpuUnpaddedBatchMode.Descriptor instead.
func (TfrtSavedModelConfig_TpuUnpaddedBatchMode) Number ¶
func (x TfrtSavedModelConfig_TpuUnpaddedBatchMode) Number() protoreflect.EnumNumber
func (TfrtSavedModelConfig_TpuUnpaddedBatchMode) String ¶
func (x TfrtSavedModelConfig_TpuUnpaddedBatchMode) String() string
func (TfrtSavedModelConfig_TpuUnpaddedBatchMode) Type ¶
func (TfrtSavedModelConfig_TpuUnpaddedBatchMode) Type() protoreflect.EnumType
type TfrtSavedModelSourceAdapterConfig ¶
type TfrtSavedModelSourceAdapterConfig struct { SavedModelConfig *TfrtSavedModelConfig `protobuf:"bytes,1,opt,name=saved_model_config,json=savedModelConfig,proto3" json:"saved_model_config,omitempty"` // contains filtered or unexported fields }
Config proto for TfrtSavedModelSourceAdapter.
func (*TfrtSavedModelSourceAdapterConfig) Descriptor
deprecated
func (*TfrtSavedModelSourceAdapterConfig) Descriptor() ([]byte, []int)
Deprecated: Use TfrtSavedModelSourceAdapterConfig.ProtoReflect.Descriptor instead.
func (*TfrtSavedModelSourceAdapterConfig) GetSavedModelConfig ¶
func (x *TfrtSavedModelSourceAdapterConfig) GetSavedModelConfig() *TfrtSavedModelConfig
func (*TfrtSavedModelSourceAdapterConfig) ProtoMessage ¶
func (*TfrtSavedModelSourceAdapterConfig) ProtoMessage()
func (*TfrtSavedModelSourceAdapterConfig) ProtoReflect ¶
func (x *TfrtSavedModelSourceAdapterConfig) ProtoReflect() protoreflect.Message
func (*TfrtSavedModelSourceAdapterConfig) Reset ¶
func (x *TfrtSavedModelSourceAdapterConfig) Reset()
func (*TfrtSavedModelSourceAdapterConfig) String ¶
func (x *TfrtSavedModelSourceAdapterConfig) String() string
type ThreadPoolFactoryConfig ¶
type ThreadPoolFactoryConfig struct { // The config proto for a ThreadPoolFactory in the ThreadPoolFactory registry. ThreadPoolFactoryConfig *anypb.Any `` /* 134-byte string literal not displayed */ // contains filtered or unexported fields }
Configuration for a thread pool factory.
func (*ThreadPoolFactoryConfig) Descriptor
deprecated
func (*ThreadPoolFactoryConfig) Descriptor() ([]byte, []int)
Deprecated: Use ThreadPoolFactoryConfig.ProtoReflect.Descriptor instead.
func (*ThreadPoolFactoryConfig) GetThreadPoolFactoryConfig ¶
func (x *ThreadPoolFactoryConfig) GetThreadPoolFactoryConfig() *anypb.Any
func (*ThreadPoolFactoryConfig) ProtoMessage ¶
func (*ThreadPoolFactoryConfig) ProtoMessage()
func (*ThreadPoolFactoryConfig) ProtoReflect ¶
func (x *ThreadPoolFactoryConfig) ProtoReflect() protoreflect.Message
func (*ThreadPoolFactoryConfig) Reset ¶
func (x *ThreadPoolFactoryConfig) Reset()
func (*ThreadPoolFactoryConfig) String ¶
func (x *ThreadPoolFactoryConfig) String() string