Create a job

A job is a complete blueprint of how to perform a step in your ML workflow, like training a model, running an evaluation, or deploying a model to an inference server. For more information, see the details of launched jobs section.

How do I create a job?

Jobs will be captured automatically from any workloads that you track with W&B if your run has associated source code. You can connect source code to your runs in the following ways:

1. Log your code as an Artifact.
2. Associate your run with a git commit.
3. Set the WANDB_DOCKER environment variable to capture a container image as a source for your job.

info

wandb>=0.13.8 is required in order to create jobs.

Job naming conventions

By default, W&B automatically creates a job name for you. The name is generated depending on how the job is created (GitHub, code artifact, or Docker image). The following table describes the job naming convention used for each job source:

Source	Naming convention
GitHub	job-<git-remote-url>-<path-to-script>
Code artifact	job-<code-artifact-name>
Docker image	job-<image-name>