Switch branches/tags
Nothing to show
Find file Copy path
e254e02 Aug 25, 2018
1 contributor

Users who have contributed to this file

65 lines (38 sloc) 2.72 KB

Submitting a job to train your model locally or in the cloud

Now that the new project is open in Visual Studio Code, you can submit a model training job to your different compute targets. (local or VM with docker).

Visual Studio Code Tools for AI provides multiple ways to submit a model training job.

  • Context Menu (right click) in Explorer view - AI: Submit Job.

    submit job

  • Context Menu (right click) in active document window - AI: Submit Job.

    submit job

  • Context Menu (right click) in AI EXPLORER - AI: Submit Job.

    submit job

  • From the command palette: "AI: Submit Job".

The typical job submitting flow consists of several steps:

  1. Select a platform (job service)

    submit job

  2. Select a configuration

    submit job

  3. Enter a job name

    By default the startup script name is used as the job name.

    submit job

  4. Confirm the job properties

Once all necessary information are collected, a json file ai_job_properties.json will be created and opened in editor window. You can review and modify the job properties here. To submit the job, click Finish button. Click Cancel to cancel the job.

submit job

By default, job property file is not always be opened for review. You can change this behavior by adding

"ai.submission.always-open-jobproperties": true 

to VS Code User Settings (CTRL+comma).

[!NOTE] The steps varies slightly depends on how you launch the command and the target platform you chose. Before you submitting the job, please open a python script file which will be used as startup script. You can confirm/change the start script setting later by editing the job property file.

Once the job is submitted, the embedded-terminal displays the progress of the runs.

Submitting Job with Docker when using Linux VM

When submitting job to linux VM, we could use docker to run the job by editing the ai_job_properties.json.

Docker Options

Please make sure the docker is installed and well-prepared on your Linux VM.

We provide some candidate docker image for you to choose, you may set "userRoot": true to avoid unexpected errors. You may also use your own dokcer image.

Please make sure the type is consistent with the image, use "type": "NvidiaDocker" if it's a docker image with NVIDIA GPU.

If the docker image does not exist on your Linux VM, it will pull the docker image from remote. Please make sure the network is available and be patient.