Add job to queue

Add launch jobs to a W&B Launch queue using the App UI or CLI to schedule ML workloads on target resources.

This page describes how to add launch jobs to a launch queue. Adding a job to a queue submits it to run on the queue’s target resource. You can schedule ML workloads against the compute environment your team configured.

Ensure that you, or someone on your team, already configured a launch queue. For more information, see Set up Launch.

Add jobs to your queue

Add jobs to your queue interactively with the W&B App or programmatically with the W&B CLI.

W&B App

To add a job to your queue interactively with the W&B App:

1. Navigate to your W&B project page.
2. Select the Jobs icon in the project sidebar:

3. The Jobs page displays a list of W&B launch jobs created from previously executed W&B runs.

4. Select the Launch button next to the job name. A modal appears.
5. From the Job version dropdown, select the version of the launch job you want to use. Launch jobs are versioned like any other W&B artifact. W&B creates different versions of the same launch job if you modify the software dependencies or source code used to run the job.
6. Within the Overrides section, provide new values for any inputs configured for your launch job. Common overrides include a new entrypoint command, arguments, or values in the wandb.Run.config of your new W&B run.

To copy and paste values from other W&B runs that used your launch job, select the Paste from… button. 7. From the Queue dropdown, select the name of the launch queue you want to add your launch job to. 8. Use the Job Priority dropdown to specify the priority of your launch job. If the launch queue doesn’t support prioritization, W&B sets the priority to Medium. 9. Optional: Follow this step only if your team admin created a queue config template. Within the Queue Configurations field, provide values for configuration options that your team admin created. For example, the following image shows a team admin who configured AWS instance types that the team can use. In this case, team members can select either the ml.m4.xlarge or ml.p3.xlarge compute instance type to train their model.

10. Select the Destination project, where the resulting run appears. This project must belong to the same entity as the queue.
11. Select the Launch now button.

Your job is now queued and runs on the queue’s target resource. The resulting run appears in the destination project you selected.

W&B CLI

Use the wandb launch command to add jobs to a queue. For example, using the script from the Quickstart guide, create a JSON file with the following hyperparameter overrides:

config.json

{
  "overrides": {
      "args": [],
      "run_config": {
          "learning_rate": 0,
          "epochs": 0
      },   
      "entry_point": []
  }
}

W&B Launch uses the default parameters if you don’t provide a JSON configuration file.

To override the queue configuration, or if your launch queue doesn’t have a configuration resource defined, specify the resource_args key in your config.json file. For example, continuing the preceding example, your config.json file might look similar to the following:

config.json

{
  "overrides": {
      "args": [],
      "run_config": {
          "learning_rate": 0,
          "epochs": 0
      },
      "entry_point": []
  },
  "resource_args": {
        "[RESOURCE-TYPE]" : {
            "[KEY]": "[VALUE]"
        }
  }
}

Replace the bracketed placeholders with your own values.

Provide the name of the queue for the queue (-q) flag, the name of the job for the job (-j) flag, and the path to the configuration file for the config (-c) flag:

wandb launch -j [JOB-NAME] -q [QUEUE-NAME] \
-e [ENTITY-NAME] -c path/to/config.json

If you work within a W&B team, specify the entity flag (-e) to indicate which entity the queue uses.

After the command succeeds, your job is added to the specified queue and runs on the queue’s target resource.