文档

create_quant_config

Function Usage

Finds all quantizable layers in a graph, creates a quantization configuration file, and writes the quantization configuration of the quantizable layers to the configuration file.

Constraints

Restrictions: Due to the data format changes, the values of quantization parameters in the generated quantization configuration file are different from those in the simplified configuration file. However, the accuracy is not affected.

Prototype

create_quant_config(config_file, model_file, weights_file, skip_layers=None, batch_num=1, activation_offset=True, config_defination=None)

Command-Line Options

Parameter

Input/Return

Description

Restriction

config_file

Input

Path and name of the quantization configuration file

The existing file (if any) in the path will be overwritten upon this API call.

A string

model_file

Input

Definition file of the Caffe model (.prototxt).

A string

weights_file

Input

Weight file of the Caffe model (.caffemodel).

A string

skip_layers

Input

Layers to skip quantizing.

Default: None

A list of strings.

Restriction: If a simplified quantization configuration file is used as the input, this parameter must be set in the configuration file. In this case, the parameter setting in the input does not take effect.

batch_num

Input

Number of batches used for quantization, that is, the number of batches used to generate quantization factors.

Type: int

Valid Value: an integer greater than or equal to 0

Default value: 1

Restrictions:

  • batch_num must not be too large. The product of batch_num and batch_size equals the number of images used during quantization. Too many images consume too much memory.
  • If a simplified quantization configuration file is used as the input, this parameter must be set in the configuration file. In this case, the parameter setting in the input does not take effect.

activation_offset

Input

Whether to quantize activations with offset.

Default: True

A bool.

Restriction: If a simplified quantization configuration file is used as the input, this parameter must be set in the configuration file. In this case, the parameter setting in the input does not take effect.

config_defination

Input

Whether to create a simplified quantization configuration file quant.cfg from the calibration_config_caffe.proto file in /amct_caffe/proto/calibration_config_caffe.proto under the AMCT installation path.

For details about the parameters in the calibration_config_caffe.proto file and the generated simplified quantization configuration file quant.cfg, see Simplified PTQ Configuration File.

Default: None

A string.

Restriction: If None, a configuration file is generated based on the remaining arguments (skip_layers, batch_num, and activation_offset). In other cases, a configuration file in JSON format is generated based on this argument.

Returns

None

Outputs

A quantization configuration file in JSON format. (When quantization is performed again, this API will overwrite the existing configuration file in the output directory.)

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
{
    "version":1,
    "batch_num":2,
    "activation_offset":true,
    "joint_quant":false,
    "do_fusion":true,
    "skip_fusion_layers":[],
    "conv1":{
        "quant_enable":true,
        "activation_quant_params":{
            "max_percentile":0.999999,
            "min_percentile":0.999999,
            "search_range":[
                0.7,
                1.3
            ],
            "search_step":0.01,
            "act_algo":"ifmr",
            "asymmetric":false
        },
        "weight_quant_params":{
            "wts_algo":"arq_quantize",
            "channel_wise":true
        }
    },
    "conv2":{
        "quant_enable":true,
        "activation_quant_params":{
            "max_percentile":0.999999,
            "min_percentile":0.999999,
            "search_range":[
                0.7,
                1.3
            ],
            "search_step":0.01,
            "act_algo":"ifmr",
            "asymmetric":false
        },
        "weight_quant_params":{
            "wts_algo":"arq_quantize",
            "channel_wise":false
        }
     }
}

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
from amct_caffe import create_quant_config
# Generate a quantization configuration file based on parameters.
create_quant_config(config_file="./configs/config.json",
                    model_file="./pretrained_model/model.prototxt",
                    weights_file="./pretrained_model/model.caffemodel",
                    skip_layers=None,
                    batch_num=1,
                    activation_offset=True)
# Generate a quantization configuration file using a simplified configuration file.
create_quant_config(config_file="./configs/config.json",
                    model_file="./pretrained_model/model.prototxt",
                    weights_file="./pretrained_model/model.caffemodel",
                    config_defination="./configs/quant.cfg")
Parent topic: PTQ APIs