Quantization

若API“是否支持”“是”“限制与说明”“-”,说明此API和原生API支持度保持一致。

API名称

是否支持

限制与说明

torch.ao.quantization.prepare_qat

-

torch.ao.quantization.convert

-

torch.ao.quantization.QuantStub

-

torch.ao.quantization.DeQuantStub

-

torch.ao.quantization.QuantWrapper

-

torch.ao.quantization.qconfig_mapping.QConfigMapping

-

torch.ao.quantization.qconfig_mapping.QConfigMapping.from_dict

-

torch.ao.quantization.qconfig_mapping.QConfigMapping.set_global

-

torch.ao.quantization.qconfig_mapping.QConfigMapping.set_module_name

-

torch.ao.quantization.qconfig_mapping.QConfigMapping.set_module_name_object_type_order

-

torch.ao.quantization.qconfig_mapping.QConfigMapping.set_module_name_regex

-

torch.ao.quantization.qconfig_mapping.QConfigMapping.set_object_type

-

torch.ao.quantization.qconfig_mapping.QConfigMapping.to_dict

-

torch.ao.quantization.qconfig_mapping.get_default_qconfig_mapping

-

torch.ao.quantization.qconfig_mapping.get_default_qat_qconfig_mapping

-

torch.ao.quantization.backend_config.BackendConfig

-

torch.ao.quantization.backend_config.BackendConfig.configs

-

torch.ao.quantization.backend_config.BackendConfig.from_dict

-

torch.ao.quantization.backend_config.BackendConfig.set_backend_pattern_config

-

torch.ao.quantization.backend_config.BackendConfig.set_backend_pattern_configs

-

torch.ao.quantization.backend_config.BackendConfig.set_name

-

torch.ao.quantization.backend_config.BackendConfig.to_dict

-

torch.ao.quantization.backend_config.BackendPatternConfig

-

torch.ao.quantization.backend_config.BackendPatternConfig.add_dtype_config

-

torch.ao.quantization.backend_config.BackendPatternConfig.from_dict

-

torch.ao.quantization.backend_config.BackendPatternConfig.set_dtype_configs

-

torch.ao.quantization.backend_config.BackendPatternConfig.set_fused_module

-

torch.ao.quantization.backend_config.BackendPatternConfig.set_fuser_method

-

torch.ao.quantization.backend_config.BackendPatternConfig.set_observation_type

-

torch.ao.quantization.backend_config.BackendPatternConfig.set_pattern

-

torch.ao.quantization.backend_config.BackendPatternConfig.set_qat_module

-

torch.ao.quantization.backend_config.BackendPatternConfig.set_reference_quantized_module

-

torch.ao.quantization.backend_config.BackendPatternConfig.set_root_module

-

torch.ao.quantization.backend_config.BackendPatternConfig.to_dict

-

torch.ao.quantization.backend_config.DTypeConfig

-

torch.ao.quantization.backend_config.DTypeConfig.from_dict

-

torch.ao.quantization.backend_config.DTypeConfig.to_dict

-

torch.ao.quantization.backend_config.DTypeWithConstraints

-

torch.ao.quantization.backend_config.ObservationType

-

torch.ao.quantization.backend_config.ObservationType.INPUT_OUTPUT_NOT_OBSERVED

-

torch.ao.quantization.backend_config.ObservationType.OUTPUT_SHARE_OBSERVER_WITH_INPUT

-

torch.ao.quantization.backend_config.ObservationType.OUTPUT_USE_DIFFERENT_OBSERVER_AS_INPUT

-

torch.ao.quantization.fx.custom_config.FuseCustomConfig

-

torch.ao.quantization.fx.custom_config.FuseCustomConfig.from_dict

-

torch.ao.quantization.fx.custom_config.FuseCustomConfig.set_preserved_attributes

-

torch.ao.quantization.fx.custom_config.FuseCustomConfig.to_dict

-

torch.ao.quantization.fx.custom_config.PrepareCustomConfig

-

torch.ao.quantization.fx.custom_config.PrepareCustomConfig.from_dict

-

torch.ao.quantization.fx.custom_config.PrepareCustomConfig.set_float_to_observed_mapping

-

torch.ao.quantization.fx.custom_config.PrepareCustomConfig.set_input_quantized_indexes

-

torch.ao.quantization.fx.custom_config.PrepareCustomConfig.set_non_traceable_module_classes

-

torch.ao.quantization.fx.custom_config.PrepareCustomConfig.set_non_traceable_module_names

-

torch.ao.quantization.fx.custom_config.PrepareCustomConfig.set_output_quantized_indexes

-

torch.ao.quantization.fx.custom_config.PrepareCustomConfig.set_preserved_attributes

-

torch.ao.quantization.fx.custom_config.PrepareCustomConfig.set_standalone_module_class

-

torch.ao.quantization.fx.custom_config.PrepareCustomConfig.set_standalone_module_name

-

torch.ao.quantization.fx.custom_config.PrepareCustomConfig.to_dict

-

torch.ao.quantization.fx.custom_config.ConvertCustomConfig

-

torch.ao.quantization.fx.custom_config.ConvertCustomConfig.from_dict

-

torch.ao.quantization.fx.custom_config.ConvertCustomConfig.set_observed_to_quantized_mapping

-

torch.ao.quantization.fx.custom_config.ConvertCustomConfig.set_preserved_attributes

-

torch.ao.quantization.fx.custom_config.ConvertCustomConfig.to_dict

-

torch.ao.quantization.fx.custom_config.StandaloneModuleConfigEntry

-

torch.ao.quantization.observer.ObserverBase

-

torch.ao.quantization.observer.ObserverBase.with_args

-

torch.ao.quantization.observer.ObserverBase.with_callable_args

-

torch.ao.quantization.observer.MinMaxObserver

-

torch.ao.quantization.observer.MinMaxObserver.calculate_qparams

-

torch.ao.quantization.observer.MinMaxObserver.forward

-

torch.ao.quantization.observer.MinMaxObserver.reset_min_max_vals

-

torch.ao.quantization.observer.MovingAverageMinMaxObserver

-

torch.ao.quantization.observer.PerChannelMinMaxObserver

-

torch.ao.quantization.observer.PerChannelMinMaxObserver.reset_min_max_vals

-

torch.ao.quantization.observer.MovingAveragePerChannelMinMaxObserver

-

torch.ao.quantization.observer.HistogramObserver

-

torch.ao.quantization.observer.PlaceholderObserver

-

torch.ao.quantization.observer.RecordingObserver

-

torch.ao.quantization.observer.NoopObserver

-

torch.ao.quantization.observer.get_observer_state_dict

-

torch.ao.quantization.observer.load_observer_state_dict

-

torch.ao.quantization.observer.default_observer

-

torch.ao.quantization.observer.default_placeholder_observer

-

torch.ao.quantization.observer.default_debug_observer

-

torch.ao.quantization.observer.default_weight_observer

-

torch.ao.quantization.observer.default_histogram_observer

-

torch.ao.quantization.observer.default_per_channel_weight_observer

-

torch.ao.quantization.observer.default_dynamic_quant_observer

-

torch.ao.quantization.observer.default_float_qparams_observer

-

torch.ao.quantization.fake_quantize.FakeQuantize

-

torch.ao.quantization.fake_quantize.FixedQParamsFakeQuantize

-

torch.ao.quantization.fake_quantize.FusedMovingAvgObsFakeQuantize

-

torch.ao.quantization.fake_quantize.disable_fake_quant

-

torch.ao.quantization.fake_quantize.enable_fake_quant

-

torch.ao.quantization.fake_quantize.disable_observer

-

torch.ao.quantization.fake_quantize.enable_observer

-

torch.ao.quantization.qconfig.QConfig

-

torch.ao.quantization.qconfig.default_qconfig

-

torch.ao.quantization.qconfig.default_debug_qconfig

-

torch.ao.quantization.qconfig.default_per_channel_qconfig

-

torch.ao.quantization.qconfig.default_dynamic_qconfig

-

torch.ao.quantization.qconfig.float16_dynamic_qconfig

-

torch.ao.quantization.qconfig.float16_static_qconfig

-

torch.ao.quantization.qconfig.per_channel_dynamic_qconfig

-

torch.ao.quantization.qconfig.float_qparams_weight_only_qconfig

-

torch.ao.quantization.qconfig.default_qat_qconfig

-

torch.ao.quantization.qconfig.default_weight_only_qconfig

-

torch.ao.quantization.qconfig.default_activation_only_qconfig

-

torch.ao.quantization.qconfig.default_qat_qconfig_v2

-

torch.ao.nn.intrinsic.LinearReLU

-

torch.ao.nn.intrinsic.qat.LinearReLU

-

torch.ao.nn.intrinsic.qat.ConvBn1d

-

torch.ao.nn.intrinsic.qat.ConvBnReLU1d

-

torch.ao.nn.intrinsic.qat.ConvBnReLU2d

-

torch.ao.nn.intrinsic.qat.update_bn_stats

-

torch.ao.nn.intrinsic.qat.freeze_bn_stats

-

torch.ao.nn.qat.Linear

-

torch.ao.nn.quantizable.LSTM

-

torch.ao.nn.quantized.dynamic.Linear

-

torch.ao.nn.quantized.dynamic.LSTM

支持fp32

torch.ao.nn.quantized.dynamic.GRU

支持fp32

torch.ao.nn.quantized.dynamic.RNNCell

支持fp32

torch.ao.nn.quantized.dynamic.LSTMCell

支持fp32

torch.ao.nn.quantized.dynamic.GRUCell

支持fp32

torch.ao.ns._numeric_suite.compare_weights

-

torch.ao.ns._numeric_suite.get_logger_dict

-

torch.ao.ns._numeric_suite.Logger

-

torch.ao.ns._numeric_suite.Logger.forward

-

torch.ao.ns._numeric_suite.ShadowLogger

-

torch.ao.ns._numeric_suite.ShadowLogger.forward

-

torch.ao.ns._numeric_suite.OutputLogger

-

torch.ao.ns._numeric_suite.OutputLogger.forward

-

torch.ao.ns._numeric_suite.Shadow

-

torch.ao.ns._numeric_suite.Shadow.forward

-

torch.ao.ns._numeric_suite.Shadow.add

支持bf16,fp16,fp32,fp64,uint8,int8,int16,int32,int64,bool

torch.ao.ns._numeric_suite.Shadow.add_scalar

支持bf16,fp16,fp32,fp64,uint8,int8,int16,int32,int64,bool

torch.ao.ns._numeric_suite.Shadow.mul

支持bf16,fp16,fp32,fp64,uint8,int8,int16,int32,int64,bool

torch.ao.ns._numeric_suite.Shadow.mul_scalar

支持bf16,fp16,fp32,fp64,uint8,int8,int16,int32,int64,bool

torch.ao.ns._numeric_suite.Shadow.cat

支持bf16,fp16,fp32,fp64,uint8,int8,int16,int32,int64,bool

torch.ao.ns._numeric_suite.Shadow.add_relu

支持bf16,fp16,fp32,uint8,int8,int32,int64

torch.ao.ns._numeric_suite.prepare_model_with_stubs

-

torch.ao.ns._numeric_suite.compare_model_stub

-

torch.ao.ns._numeric_suite.get_matching_activations

-

torch.ao.ns._numeric_suite.prepare_model_outputs

-

torch.ao.ns._numeric_suite.compare_model_outputs

-

torch.ao.ns._numeric_suite_fx.OutputLogger

-

torch.ao.ns._numeric_suite_fx.OutputLogger.forward

-

torch.ao.ns._numeric_suite_fx.OutputComparisonLogger

-

torch.ao.ns._numeric_suite_fx.OutputComparisonLogger.forward

-

torch.ao.ns._numeric_suite_fx.NSTracer

-

torch.ao.ns._numeric_suite_fx.NSTracer.is_leaf_module

-

torch.ao.ns._numeric_suite_fx.extract_weights

-

torch.ao.ns._numeric_suite_fx.add_loggers

-

torch.ao.ns._numeric_suite_fx.extract_logger_info

-

torch.ao.ns._numeric_suite_fx.add_shadow_loggers

-

torch.ao.ns._numeric_suite_fx.extract_shadow_logger_info

-

torch.ao.ns._numeric_suite_fx.extend_logger_results_with_comparison

-

torch.ao.ns._numeric_suite_fx.prepare_n_shadows_model

-

torch.ao.ns._numeric_suite_fx.loggers_set_enabled

-

torch.ao.ns._numeric_suite_fx.loggers_set_save_activations

-

torch.ao.ns._numeric_suite_fx.convert_n_shadows_model

-

torch.ao.ns._numeric_suite_fx.extract_results_n_shadows_model

-

torch.ao.ns._numeric_suite_fx.print_comparisons_n_shadows_model

-

torch.ao.ns.fx.utils.compute_sqnr

-

torch.ao.quantization.get_default_qat_qconfig

说明:

该API为Quantization Aware Training依赖接口,未在PyTorch对外文档中呈现,具体使用请参考LINK

在aarch64系统中只能使用qnnpack backend