Help Center

A

• A

  AI Engine

  An AI engine is a framework that allows you to develop machine learning and deep learning model training jobs, such as TensorFlow and MXNet.

  AK/SK

  The access key ID (AK) and secret access key (SK) are used to access OBS buckets on ModelArts.

B

• B

  Batch Service

  The batch services are used to process batch data stored in OBS buckets at a time.

  Boot File

  A training job may have multiple code files. A boot file specifies the file from which the training job starts to run.

  Built-in Algorithms

  Built-in algorithms are training code built in ModelArts that can be directly reused.

C

• C

  Code Directory

  A code directory is the OBS path (usually a directory) where the code corresponding to a training job is located.

D

• D

  Dataset

  A dataset is sample data stored in an OBS bucket and used for training models.ModelArts can manage the versions of datasets and switch different versions in different scenarios.

  Development Environment

  A development environment is an editor required for developing AI training jobs.

E

• E

  Edge Service

  A model can be deployed as a service on an edge server, then the edge server provides the service to peripheral devices.

  Exeml

  Auto Learning is the process of automating model design, parameter tuning and training, and model compression and deployment with the labeled data. The process is free of coding and does not require developers' experience in model development.

J

• J

  Job Parameters

  Job parameters are the running parameters specified when a training job is submitted. You can save complex parameters and reuse them during subsequent job creation.

M

• M

  Metamodel

  A metamodel is a model of a model.

N

• N

  Notebook

  Notebook is an online interactive code development and debugging tool powered on the open source Jupyter Notebook. It is widely used in the AI field.

O

• O

  Online Service

  A model can be deployed as a cloud service. You can directly access the service by calling the RESTful API, which is used for the inference of a single piece of data.

R

• R

  Running Parameters

  Running parameters are dynamic parameters specified when a training job is created. The dynamic parameters are received and processed by the code of the training job. You can specify the parameter values as required.

T

• T

  Tracing Graph

  Data objects operated by you during the development of an AI service form a directed graph. ModelArts records the complete graph and displays the graph information on the management console, allowing for accurate track on the entire development process.

  Training Job

  A training job is a task submitted by you to train a model. You can edit and develop the code logic of the task in the development environment. After job running, a model is outputted.

V

• V

  Version Management

  In ModelArts, multiple versions of data are generated when data (such as datasets, training jobs, and models) of jobs of different types is modified. ModelArts records version information and displays it on the management console for you to accurately trace and compare changes between versions.

W

• W

  Weight

  If multiple models need to be deployed in an online service to provide services, you can configure different weights for each model. The online service forwards request traffic to each model based on weights.