huawei-npu

Introduction

huawei-npu is a management add-on for Huawei NPU devices in containers.

After this add-on is installed, you can create Ascend-accelerated nodes to quickly and efficiently process inference and image recognition.

Constraints

If Ascend-accelerated nodes are used in a cluster, the huawei-npu add-on must be installed.
After an AI-accelerated node is migrated, the node will be reset. Manually reinstall the NPU driver.

Installing the Add-on

Log in to the CCE console and access the cluster console. Choose Add-ons in the navigation pane, locate huawei-npu on the right, and click Install.

Set NPU parameters. The add-on uses the following parameters by default. The default NPU settings provided by the add-on can satisfy most scenarios and require no changes.

{
	"check_frequency_failed_threshold": 100,
	"check_frequency_fall_times": 3,
	"check_frequency_gate": false,
	"check_frequency_recover_threshold": 100,
	"check_frequency_rise_times": 2,
	"container_path": "/usr/local/HiAI_unused",
	"host_path": "/usr/local/HiAI_unused"
}