Learning about AIACC-ACSpeed | Install and Use AIACC-ACSpeed

This article describes how to install and use AIACC-ACSpeed V1.1.0.

AIACC-ACSpeed is a training accelerator designed to optimize communication between distributed nodes in distributed training scenarios. AIACC-ACSpeed leverages optimized, decoupled modules to optimize distributed training jobs in terms of compatibility, applicability, and performance. This article describes how to install and use AIACC-ACSpeed V1.1.0.

Prerequisites

An Alibaba Cloud GPU-accelerated instance that meets the following requirements is created:

The OS is Alibaba Cloud Linux, CentOS 7.x and later, or Ubuntu 16.04 and later.
An NVIDIA driver and CUDA 10.0 or later are installed.

Supported Versions

AIACC-ACSpeed (ACSpeed) V1.1.0 supports PyTorch, CUDA, Python, and NGC images. The following table describes the supported versions.

TABLE

Install AIACC-ACSpeed

1. Run the following command to download ACSpeed V1.1.0:

wget https://ali-perseus-release.oss-cn-huhehaote.aliyuncs.com/ACSpeed/acspeed-1.1.0.tar.gz

2. Run the following command to install ACSpeed V1.1.0:

pip install acspeed-1.1.0.tar.gz

Use AIACC-ACSpeed

You can use ACSpeed by importing the ACSpeed module into your code.

Add a line of import command to the file corresponding to the main function of the training code to import ACSpeed code. You can add it to the place where torch is imported. The following commands provide an example:

import torch
import acspeed

0 1 0

Share on

Community

Learning about AIACC-ACSpeed | Install and Use AIACC-ACSpeed

Prerequisites

Supported Versions

Install AIACC-ACSpeed

Use AIACC-ACSpeed

Read previous post:

Read next post:

Alibaba Cloud Community

You may also like

Comments

Alibaba Cloud Community

Related Products

Container Compute Service (ACS)

Container Service for Kubernetes

GPU(Elastic GPU Service)

EasyDispatch for Field Service Management