--calibration_config

Function Usage

Sets the path (including the file name) of the simplified configuration file for PTQ.

Include this option if you have created a simplified configuration file for PTQ (that is, you control which layers to be quantized or not to be quantized).

See Also

This option is mutually exclusive with --batch_num.

Argument

Argument: path (including the file name) of the simplified configuration file.

Remarks: The path (including the file name) can contain letters, digits, underscores (_), hyphens (-), and periods (.).

Default: none.

Suggestions and Benefits

None

Examples

--calibration_config=./config.cfg

The following is an example of the file. The Caffe framework is used as an example. For details about the parameters, see. (The arguments used in the following are for reference only. Modify them as required.)

# global quantize parameter
batch_num : 2
activation_offset : true
joint_quant : false
skip_layers : "conv_1"
skip_layer_types:"Convolution"
do_fusion: true
skip_fusion_layers : "conv_1"
common_config : {
    arq_quantize : {
        channel_wise : true
    }
    ifmr_quantize : {
        search_range_start : 0.7
        search_range_end : 1.3
        search_step : 0.01
        max_percentile : 0.999999
        min_percentile : 0.999999
    }
}
 
override_layer_types : {
    layer_type : "InnerProduct"
    calibration_config : {
        arq_quantize : {
            channel_wise : false
        }
        ifmr_quantize : {
            search_range_start : 0.8
            search_range_end : 1.2
            search_step : 0.02
            max_percentile : 0.999999
            min_percentile : 0.999999
        }
    }
}

Constraints

This option is used only in PTQ scenarios.