ModelEvaluator

Applicability

Product	Supported
Atlas 350 Accelerator Card	√
Atlas A3 training product/Atlas A3 inference product	√
Atlas A2 training product/Atlas A2 inference product	√
Atlas 200I/500 A2 inference product	√
Atlas inference product	√
Atlas training product	√

Description

Provides a Python instance based on the input data of the bin type for a model to calibrate and infer the model.

Prototype

class ModelEvaluator(AutoCalibrationEvaluatorBase):
def __init__(self, data_dir, input_shape, data_types):

Parameters

Parameter	Input/Output	Description
data_dir	Input	Path of the .bin dataset that matches the model. A string. Format: "data/input1/;data/input2/" Restrictions: The path can contain letters, digits, underscores (_), hyphens (-), and periods (.). If the model has multiple inputs and each input has multiple batches of data, these batches must be stored in different directories and the file in the directories must be arranged in ascending order by name. All input paths must be enclosed in double quotation marks ("") and separated with semicolons (;). The shape of arrays stored in a single .bin file must match that specified by input_shape. For example, if the shape of arrays stored in a single .bin file is 1 × 224 × 224 × 3, that specified in input_shape must also be 1 × 224 × 224 × 3. If multiple .bin files are required for quantization, you can adjust the batch number by using batch_num.
input_shape	Input	Input shape A string. Format: "input_name1:n1,c1,h1,w1;input_name2:n2,c2,h2,w2" Restrictions: The specified nodes must be enclosed in double quotation marks ("") and separated with semicolons (;).
data_types	Input	Input data type. A string. Format: "float32;float64" Restrictions: If a model has multiple inputs of different data types, you need to specify all the involved data types. All the data types must be enclosed in double quotation marks ("") and separated with semicolons (;) in the input node sequence.

Parameter

Input/Output

Description

data_dir

Input

Path of the .bin dataset that matches the model.

A string.

Format: "data/input1/;data/input2/"

Restrictions:

The path can contain letters, digits, underscores (_), hyphens (-), and periods (.).
If the model has multiple inputs and each input has multiple batches of data, these batches must be stored in different directories and the file in the directories must be arranged in ascending order by name. All input paths must be enclosed in double quotation marks ("") and separated with semicolons (;).
The shape of arrays stored in a single .bin file must match that specified by input_shape. For example, if the shape of arrays stored in a single .bin file is 1 × 224 × 224 × 3, that specified in input_shape must also be 1 × 224 × 224 × 3. If multiple .bin files are required for quantization, you can adjust the batch number by using batch_num.

input_shape

Input

Input shape

A string.

Format: "input_name1:n1,c1,h1,w1;input_name2:n2,c2,h2,w2"

Restrictions: The specified nodes must be enclosed in double quotation marks ("") and separated with semicolons (;).

data_types

Input

Input data type.

A string.

Format: "float32;float64"

Restrictions: If a model has multiple inputs of different data types, you need to specify all the involved data types. All the data types must be enclosed in double quotation marks ("") and separated with semicolons (;) in the input node sequence.

Returns

Returns a Python instance.

Example

import amct_pytorch as amct

evaluator = amct.ModelEvaluator(
    data_dir="./data/input_bin/", 
    input_shape="input:32,3,224,224", 
    data_types="float32")

Parent topic: Common API