运行和监控作业 (Run:ai CLI)#

本节介绍如何使用 Run:ai CLI 提交和检查作业状态。

  1. 使用 login 选项登录 runai。

    运行以下命令并输入用户名和密码。

    runai login

    示例结果

    1researcher1@basepod-head1:~/runai_dir$ runai login
2Username: researcher1@nvidia.com
3Password:
4INFO[0022] Logged in successfully

  2. 提交分布式训练作业。

    runai submit-mpi dist-job1 -i \n gcr.io/run-ai-demo/quickstart-distributed:tf-2.1.0 -g 4 -p researcher1

    示例结果

    1researcher1@basepod-head1:~/runai_dir$ runai submit-mpi dist-job1 -i \n gcr.io/run-ai-demo/quickstart-distributed:tf-2.1.0 -g 4 -p researcher1
2Job dist-job1 submitted successfully.

  3. 使用 runai describe 检查作业状态。

    runai describe job dist-job1 -p researcher1

    示例结果

     1researcher1@basepod-head1:~/runai_dir$ runai describe job dist-job1 -p researcher1
 2Name: dist-job1
 3Namespace: runai-researcher1
 4Type: Train
 5Status: Running
 6Duration: 5m0
 7GPUs: 0
 8Total Requested GPUs: 0
 9Allocated GPUs: 2
10Allocated GPUs memory: 0
11Running PODs: 3
12Pending PODs: 0
13Parallelism: 1
14Completions: 1
15Succeeded PODs: 0
16Failed PODs: 0
17Is Distributed Workload: true
18Service URLs:
19Command Line: runai submit-mpi dist-job1 --processes=2 -g 1 -i gcr.io/run-ai-demo/quickstart-distributed:tf-2.1.0 -e RUNAI_SLEEP_SECS=60