AI Training - Getting started

Last updated 18th May, 2021.

Objective

This guide covers the initialisation of AI Training and the submission of jobs through the OVHcloud Control Panel.

Requirements

• a Public cloud project
• optionally container objects to attach data to the job at step 6, see our create data container guide.
• access to the OVHcloud Control Panel

Instructions

Step 1 - Going to the AI Training menu

Log in to the OVHcloud Control Panel, go to the Public Cloud section and then to the AI Training section which is located under AI & Machine Learning.

Step 2 - Activating AI Training service

Once you have read the general information and validated this services's contract terms, you can start submitting your jobs. Upon activating the AI Training service you grant OVHcloud access to your Object Storage containers. This access is only used to synchronise your data within AI Training with your containers.

Dashboard

Once AI Training is activated you land on the dashboard service with several components.

• Information: useful information for service usage
• Usage: brief summary of number of jobs according to their status
• Billing: an overview of your ongoing billing
• AI Training users: list of users that can use this service
• Jobs: list of active jobs

Step 3 - Starting a job submission

From the jobs list in the dashboard you can start the job submission by clicking the Launch a new Job button.

Step 4 - Selecting a region for your job

Each job is executed in an OVHcloud region. Each region has its own AI Training cluster with potentially varying capabilities. For more information see the capabilities. Select the desired region and click Next.

Step 5 - Providing a Docker image

A job is basically a Docker container that is run within the OVHcloud infrastructure. You need to provide a Docker image to be executed. There are several options you can choose from:

Preset Images

OVHcloud provides a set of images from which you can choose to ease the submission of your first jobs. Provided images are essentially a JupyterLab environment bundled with some Deep Learning technology such as Tensorflow or MXNet.

Custom Images

Preset images cannot cover all your needs so you can specify your own image if necessary. You can use any image that is accessible from AI Training.

This includes public images (e.g. Dockerhub), images within the shared registry or images in your added private registry. For more information, see how to add a private registry.

Once your image is chosen, click Next.

Step 6 - Attaching data to your job (optional)

You can attach data objects to your job either as input for your training workload or as output for your results (e.g. model weights).

Before attaching a data object you need to create one. A data object cannot be attached to a running job.

To attach a data object, just select from the list on the left. Next to each data, within the parenthesis, you can check the mount path in the Docker container for the submitted job. If you wish to customise this mount path, you will need to use the ovhai CLI, its installation procedure is available here.

Once the data is configured click Next.

Step 7 - Overriding the Docker entrypoint (optional)

The Docker image you provided in Step 5 includes an entrypoint for your container. You can override this entrypoint by specifying your own command. Once the entrypoint is set up click Next.

Step 8 - Specifying the amount of resources

In this step you can either select the amount of GPUs or CPUs you need for your training workload.

The max amount of GPUs or CPUs you can select for your job is region dependent. If you choose a GPU a fixed ratio of CPU is applied based on the number of GPUs. Similarly, there is a fixed ratio of Memory based on the number of CPUs. For more information see the capabilities.

Once the amount of resources is set you can see a preview of the billing rate. Click Next.

Step 9 - Submitting your job

In the final step you get an overview of the job you configured before submission. You also get the equivalent command to use with the ovhai CLI.

The AI Training service is mainly supposed to be used through the ovhai CLI. The OVHcloud Control Panel only offers a subset of the features and is meant to help you get started before using the CLI.

Finally click Submit to submit your job to the cluster.

Step 10 - Consulting your job

Once the job is submitted you are redirected to the jobs list page.

From this list you can access your job details either by clicking on its ID or by clicking on ... and selecting Details. The details include several components:

• Job Information: basic information on the job you submitted
• Container: describes the status of your job and provides you with the URL to access any service exposed by your job on the port 8080. The URL is of the form https://<JOB-ID>.job.<REGION>.training.ai.cloud.ovh.net/. If the service is not exposed on the port 8080 it is still accessible by specifying the port in the URL this way: https://<JOB-ID>-<PORT>.job.<REGION>.training.ai.cloud.ovh.net/. You can check the list of available ports in the capabilities.
• Resources: a summary of the resources consumed by the job
• Actions: available actions
• Data: list of data objects attached to the job

Step 11 - Cancelling your job

If you are done using your job, if your model converged prematurely or if you just wish to interrupt your job you can do so from the jobs list.

From the list of jobs you can list the available actions at the far right of each entry and interrupt the job by clicking Stop. Alternatively, from the job details you can also interrupt the job from the list of actions.

Feedback

Please send us your questions, feedback and suggestions to improve the service:

• On the OVHcloud Discord server

---

---

Inne przewodniki, które mogą Cię zainteresować...