Integrations / Ray

RunExperimental

Run single-node Ray workloads on Badgr

Run a Ray script on one GPU machine as a finite job. Multi-node Ray clusters are not covered.

A setup guide only. Badgr has not tested this path, so expect to adjust it.

Why use Badgr for Ray

Badgr interface
Run
Where Ray fits
Compute framework
Why Badgr
Compatible Ray jobs
  • Run a single-node Ray script on a rented GPU machine without setting up a cluster.
  • A finite job with a cost cap suits sweeps and batch work that end by themselves.
  • Badgr is clear about the limit: no multi-node Ray clusters.

What you can do

  • Parallel data processing on one large machine
  • Ray Tune sweeps that fit on a single node
  • Ray Data batch inference on one GPU

Quickstart: Run a Ray script

badgr run . --cmd "python ray_job.py" --max-cost 5

How it works

  1. 1.You run one command
  2. 2.Badgr checks it and picks a GPU
  3. 3.Your code runs and logs stream back
  4. 4.Outputs are saved and billing stops

What you get

  • Logs while it runs
  • Output files, downloaded with badgr artifacts <id>
  • A receipt with cost and duration

Requirements

  • The Badgr CLI installed and logged in (npm install -g badgr-cli)
  • A funded Badgr balance
  • A script that starts Ray itself (ray.init() on the local machine)
  • ray listed in requirements.txt

What it costs

  • A job is billed for the time it runs and stops billing when your command exits.
  • Pass --max-cost to set the most a job can spend. Badgr stops it if it reaches the cap.
  • Badgr shows the rate for the GPU it picked before it starts. This page does not quote a fixed price.

See pricing for how billing works.

Proof

No tested run is recorded for this integration yet. A receipt and example output will be added here after a real run, not before.

Troubleshooting

  • Need many machines

    This path runs on one machine. Badgr does not start a Ray cluster across several machines.

  • ray.init cannot find a cluster

    Call ray.init() with no address so Ray starts locally inside the job.

  • Out of memory

    Reduce num_cpus or num_gpus per task, or ask for a larger GPU.

Developer reference

Ready to start with Ray?

Every run or service has a spending limit you set before it starts.