Launch Job
Push a WandB Launch job for a branch to a queue for execution on GPU agents.
Usage
bash
python .claude/skills/launch-job/launch_job.py \
--branch <branch-name> \
--queue <queue-name> \
--project <wandb-project> \
[--entity <wandb-entity>] \
[--config '{"learning_rate": 0.001}'] \
[--entry-point main.py] \
[--resource kubernetes] \
[--docker-image <image>] \
[--priority 1] \
[--dry-run]
When to use
Use this skill when:
- •A branch has been implemented and is ready for training/evaluation
- •You need to queue an experiment on a GPU agent
- •Re-launching a job after fixing bugs on a branch
Notes
- •The branch must exist locally and be pushed to the remote
- •Branch name and commit hash are automatically injected into
wandb.config - •Jobs are temporary and not reusable (per project convention)
- •Use
--dry-runto verify config before launching - •The
--configflag accepts a JSON string or path to a JSON file
Arguments
$ARGUMENTS