Before You Start

MindCluster support containerized deployment, fault rescheduling, and elastic scaling of MindIE Motor by generating inference jobs of acjob type.

This section only describes the principles of related features and provides corresponding configuration examples. The provided YAML examples are not sufficient to complete the deployment of MindIE jobs. For details about the complete deployment process of MindIE Motor, see the MindIE Motor Development Guide.

Prerequisites

Before deploying MindIE Motor, ensure that the relevant components have been installed. If they are not installed, refer to the Installation and Deployment section for instructions.

Volcano
Ascend Device Plugin
Ascend Docker Runtime
Ascend Operator; the startup parameter enableGangScheduling must be set to true
ClusterD
NodeD

Supported Product Forms

Atlas 800I A2 inference server
Atlas 800I A3 SuperPoD server

Usage Methods

MindCluster supports containerized deployment, fault rescheduling, and elastic scaling of MindIE Motor through the following two methods. This section only describes the method of using the command line.

Using the command line: Deploy jobs through configured YAML files.
Using after integration: Integrate the cluster scheduling components into an existing third-party AI platform or an AI platform developed based on these components.