Launch terms and concepts

Learn key W&B Launch concepts including jobs, queues, target resources, agents, and agent environments.

This page defines the core terms and concepts used throughout the W&B Launch documentation. Use it as a reference when you encounter unfamiliar terms in setup guides, tutorials, or the W&B App.

With W&B Launch, you enqueue jobs onto queues to create runs. Jobs are Python scripts instrumented with W&B. Queues hold a list of jobs to execute on a target resource. Agents pull jobs from queues and execute them on target resources. W&B tracks Launch jobs similarly to how it tracks runs.

The following sections describe each of these concepts in more detail.

Launch job

A Launch job is a specific type of W&B Artifact that represents a task to complete. For example, common Launch jobs include training a model or triggering a model evaluation. Job definitions include:

• Python code and other file assets, including at least one runnable entrypoint.
• Information about the input (config parameter) and output (metrics logged).
• Information about the environment (for example, requirements.txt, base Dockerfile).

The three main kinds of job definitions are:

Job types	Definition	How to run this job type
Artifact-based (or code-based) jobs	Code and other assets are saved as a W&B artifact.	To run artifact-based jobs, you must configure the Launch agent with a builder.
Git-based jobs	Code and other assets are cloned from a certain commit, branch, or tag in a git repository.	To run git-based jobs, you must configure the Launch agent with a builder and git repository credentials.
Image-based jobs	Code and other assets are baked into a Docker image.	To run image-based jobs, you might need to configure the Launch agent with image repository credentials.

Although Launch jobs can perform activities not related to model training (for example, deploying a model to a Triton inference server), all jobs must call wandb.init() to complete successfully. This creates a run for tracking purposes in a W&B workspace.

Find jobs you created in the W&B App under the Jobs tab of your project workspace. From there, you can configure jobs and send them to a Launch queue to execute on different target resources.

Launch queue

Launch queues are ordered lists of jobs to execute on a specific target resource. Launch queues are first-in, first-out (FIFO). No practical limit applies to the number of queues you can have, but a good guideline is one queue per target resource. You can enqueue jobs with the W&B App UI, W&B CLI, or Python SDK. You can then configure one or more Launch agents to pull items from the queue and execute them on the queue’s target resource.

Target resources

The compute environment that a Launch queue executes jobs on is called the target resource.

W&B Launch supports the following target resources:

• Docker
• Kubernetes
• AWS SageMaker
• Google Cloud Vertex AI

Each target resource accepts a different set of configuration parameters called resource configurations. Resource configurations take on default values defined by each Launch queue, but each job can override them independently. See the documentation for each target resource for more details.

Launch agent

Launch agents are lightweight, persistent programs that periodically check Launch queues for jobs to execute. When a Launch agent receives a job, it first builds or pulls the image from the job definition, then runs it on the target resource.

One agent can poll multiple queues. However, you must configure the agent properly to support all the backing target resources for each queue it polls.

Launch agent environment

The agent environment is the environment where a Launch agent is running, polling for jobs.

The agent’s runtime environment is independent of a queue’s target resource. In other words, you can deploy agents anywhere as long as you configure them sufficiently to access the required target resources.