factor technologies
Gyms

Connect a gym to your cluster

Train a private gym on your own Kubernetes cluster. Your administrator prepares the trainer and gym images and the data and output volumes. Studio does not turn an arbitrary wheel into a training program.

You need a private workspace gym and a cluster administrator who can install the agent and edit its local training profiles.

Connect the cluster#

  1. In Kubernetes clusters, choose Connect a cluster, register a name and copy the one-time credential.
  2. Have your administrator store the credential as token in the cluster's gft-byoc-credential Secret. Request the agent installation package from your GFT contact if your organization does not already have it.
  3. Install the agent in a dedicated namespace using your approved image digests, GPU runtime and encrypted input and output volumes. The trainer needs its model and dependencies locally.
  4. Wait for the agent to appear online in Studio before choosing a training profile.

Bind the gym to a profile#

  1. In Kubernetes clusters, select the connected agent.
  2. Copy the exact environment version ID from your gym.
  3. In the matching local profile, set environment_version_id to that ID. Optionally, set model_label to a model name that is safe to share.
  4. Make sure the profile's gym image and trainer implement that version, then restart the agent.

Studio receives only the binding and the model label. Commands, image paths and volume names stay in your cluster.

Start training#

Open Train with this gym, choose the matching online profile, review the limits and confirm.

Marketplace gyms cannot run on cluster profiles: owning a private gym does not give permission to run purchased gym code without metering.

Next steps#