Scroll to navigation

DLMODEL_SOURCE(1) User Commands DLMODEL_SOURCE(1)

NAME

dlmodel_source - create a structured deep learning model directory

SYNOPSIS

dlmodel_source

DESCRIPTION

dlmodel_source helps to create a structured deep learning (DL) model directory via CLI. Please refer to the specification section below for more details.

The structured directory will be created in /tmp.

Questions for collecting model information:

Package name
The name of the DL package, e.g. inception.
Package version
The version of the DL package, e.g. 3.
Modle filepath
Absolute or relative path of the model file.
Label filepath
Absolute or relative path of the label file.
Config name
Name of the config data source (see config filepath below). If there is not any config, leave it blank.
Config filepath
Absolute or relative path of the config file.
Inference engine
The inference engine supporting the DL model's format.

MODEL DIRECTORY STRUCTURE

Example of a structured DL model directory:


<modelname-version>
|-- <model-file>
|-- <label-file>
|-- <optional-config-files>
|-- LICENSE # optional currently
`-- meta.json

METADATA FORMAT OF MODEL PACKAGE

Metadata, meta.json, describes all the details in the structured model directory.

Example of meta.json


# Note: model directory name is fight-detection-1.0.0
{
"name": "fight-detection",
"version": "1.0.0",
"inference-engine": "tensorflow",
"model": "model.pb",
"label": "labels.txt",
# optional configs
"config": {
"<optional-key>": "<optional-value>",
"<optional-key>": "<optional-value>",
...
},
"checksums-sha256": {
"model.pb": "<sha256sum>",
"labels.txt": "<sha256sum>",
"<optional-file-path>": "<sha256sum>",
...
}
}

SEE ALSO

dlmodel2deb(1)

October 2017 DLMobelBox