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.
Context Menu (right click) in active document window - AI: Submit Job.
Context Menu (right click) in AI EXPLORER - AI: Submit Job.
From the command palette: "AI: Submit Job".
The typical job submitting flow consists of several steps:
Select a platform (job service)
Select a configuration
Enter a job name
By default the startup script name is used as the job name.
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.
By default, job property file is not always be opened for review. You can change this behavior by adding
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
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
"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.