Jobs: run work that finishes

What is Jobs?

A job runs your command on a GPU, streams logs, stores the outputs and stops when the command exits. Billing stops with it. Use Jobs when the work has an end.

  • Python or Node scripts from a local folder or a GitHub repo
  • Batch inference and embedding generation
  • Fine-tuning and LoRA training
  • ComfyUI, image, video and audio transcription runs
  • Existing Slurm batch scripts via badgr sbatch

Run your first job

npm install -g badgr-cli
badgr login
badgr run . --cmd "python train.py" --gpu A100 --max-cost 10

Badgr zips the current folder, picks a runner, installs dependencies from requirements.txt or package.json and runs your command.

What happens

  1. 1.Badgr picks capacity that matches the GPU you asked for.
  2. 2.The machine starts and your project is loaded into a container.
  3. 3.Your command runs and logs stream to your terminal.
  4. 4.When the command exits, the machine is stopped and billing ends.

Cost control

  • --max-cost stops the job when spend reaches that amount. If you omit it, Badgr computes a conservative default from the GPU tier and runtime.
  • badgr down <id> stops a job at any time.
  • badgr status shows what is running and billing right now.

Get results

  • badgr logs <id> returns output from a running or finished job.
  • Outputs are stored for 48 hours; fetch them with badgr artifacts <id>.
  • badgr receipts shows cost, route, retry and teardown history for run and serve actions.

What this does not cover

  • Multi-node distributed training is not supported.
  • Spot or preemptible capacity is not offered.

Next steps