Simplified QAT Configuration File

Table 1 describes the fields in the retrain_config_caffe.proto file. Find the file in /amct_caffe/proto/retrain_config_caffe.proto in the AMCT installation path.

**Table 1** Parameter description
Message	Required	Type	Field	Description
AMCTRetrainConfig	-	-	-	Simplified QAT configuration of AMCT.
	repeated	string	skip_layers	Layers to skip. It is globally effective, which efficiently realizes the same functionality for extended features. If this field is set, you can skip the settings of the extended quant_skip_layers and xxx_skip_layers. If both skip_layers and quant_skip_layers are set, their union is used.
	repeated	string	skip_layer_types	Layers to be skipped by layer type (not supported in the current version). It is globally effective, which efficiently realizes the same functionality for extended features. If this field is set, you can skip the settings of the extended quant_skip_types and xxx_skip_types. If both skip_layer_types and quant_skip_types are set, their union is used.
	repeated	RetrainOverrideLayer	override_layer_configs	Rewrite layers by layer name, that is, perform differentiated compression on the layers. For example, if the initial upper and lower limits configured by using the global quantization configuration parameter are [-0.6, 0.6], differentiated quantization may be performed on some layers by using this parameter, and this parameter may be configured as [-0.3, 0.3]. Parameter Priority Quantization scenario: override_layer_configs>override_layer_types>retrain_data_quant_config/retrain_weight_quant_config Sparse scenario: override_layer_configs>override_layer_types>prune_config
	repeated	RetrainOverrideLayerType	override_layer_types	Rewrite layers by layer type, that is, perform differentiated compression on the layers. For example, if the initial upper and lower limits configured by using the global quantization configuration parameter are [-0.6, 0.6], differentiated quantization may be performed on some layers by using this parameter, and this parameter may be configured as [-0.3, 0.3]. Parameter Priority Quantization scenario: override_layer_configs>override_layer_types>retrain_data_quant_config/retrain_weight_quant_config Sparse scenario: override_layer_configs>override_layer_types>prune_config
	optional	uint32	batch_num	Batch count used for quantization.
	required	RetrainDataQuantConfig	retrain_data_quant_config	Quantization configuration parameter for QAT data. Global quantization configuration parameter. Parameter priority: override_layer_configs > override_layer_types > retrain_data_quant_config/retrain_weight_quant_config
	required	RetrainWeightQuantConfig	retrain_weight_quant_config	Weight quantization configuration parameter for QAT. Global quantization configuration parameter. Parameter priority: override_layer_configs > override_layer_types > retrain_data_quant_config/retrain_weight_quant_config
	repeated	string	quant_skip_layers	Skips layers that do not need to be quantized by layer name. Parameter used in the quantization scenario. If both skip_layers and quant_skip_layers are set, their union is used.
	repeated	string	quant_skip_types	Skips layers that do not need to be quantized by layer type (not supported in the current version). Parameter used in the quantization scenario. If both skip_layer_types and quant_skip_types are set, their union is used.
RetrainDataQuantConfig	-	-	-	Quantization parameter configuration for QAT data.
RetrainDataQuantConfig	-	ULQuantize	ulq_quantize	Activation quantization algorithm. Currently, only ULQ is supported.
ULQuantize	-	-	-	ULQ quantization algorithm configuration. For details about the algorithm, see ULQ Algorithm for Activation Quantization.
	optional	ClipMaxMin	clip_max_min	Initial upper and lower bounds. IFMR is used for initialization by default.
	optional	bool	fixed_min	Whether to fix the lower bound at 0. Set to true for ReLU or false for other algorithms.
ClipMaxMin	-	-	-	Initial upper and lower bounds.
	required	float	clip_max	Initial upper bound.
	required	float	clip_min	Initial lower bound.
RetrainWeightQuantConfig	-	-	-	Weight quantization parameter configuration for QAT.
RetrainWeightQuantConfig	-	ARQRetrain	arq_retrain	ARQ algorithm for weight quantization. Currently, only ARQ is supported.
ARQRetrain	-	-	-	ARQ algorithm configuration. For details about the algorithm, see ARQ Algorithm.
ARQRetrain	required	bool	channel_wise	Channel-wise ARQ enable.
RetrainOverrideLayer	-	-	-	Layer overriding configuration.
	required	string	layer_name	Layer name.
	required	RetrainDataQuantConfig	retrain_data_quant_config	Activation quantization configuration to apply.
	required	RetrainWeightQuantConfig	retrain_weight_quant_config	Weight quantization configuration to apply.
RetrainOverrideLayerType	-	-	-	Types of layers to override.
	required	string	layer_type	Layer type.
	required	RetrainDataQuantConfig	retrain_data_quant_config	Activation quantization configuration to apply.
	required	RetrainWeightQuantConfig	retrain_weight_quant_config	Weight quantization configuration to apply.

The following is an example simplified QAT configuration file (quant.cfg):

# global quantize parameter
 retrain_data_quant_config: {
     ulq_quantize: {
         clip_max_min: {
             clip_max: 6.0
             clip_min: -6.0
         }
     }
 }

 retrain_weight_quant_config: {
     arq_retrain: {
         channel_wise: true
     }
 }

 skip_layers: "conv_1"

 override_layer_types : {
     layer_type: "InnerProduct"
     retrain_weight_quant_config: {
         arq_retrain: {
            channel_wise: false
         }
     }
 }

 override_layer_configs : {   
     layer_name: "fc_5"   
     retrain_weight_quant_config: {        
         arq_retrain: {   
            channel_wise: false
         }    
      }
}

Parent topic: See Also