Skip to main content
Publishing deploys your model version to Reactor’s GPU cloud. Once published, your model is live and accepting sessions in under 3 minutes.

​
Pre-publish checklist

Before running model publish, confirm:
  • Model is registered (model register)
  • Weights are uploaded and status is completed for your target version
  • Docker image is pushed to your container registry
  • Model runs successfully in local Docker (see Local Testing)

​
Publish

reactor model publish --model my-model --version v1
The CLI validates that the required weight version and Docker image exist, then triggers deployment.

​
What happens after publishing

  1. Provisioning - Reactor allocates GPU nodes and pulls your Docker image and weights.
  2. Startup - Your model container starts and runs load().
  3. Live - Once healthy, your model is routed and ready to accept client sessions.
This typically completes in under 3 minutes. You can watch it happen in real time: 
reactor model status --model my-model --watch

​
Check deployment status

reactor model status --model my-model
StatusMeaning
registeredModel is registered but not yet published
deployingImage and weights are being pulled to GPU nodes
startingContainer is running, model is loading
liveRunning in production, accepting sessions
failedDeployment failed. Check logs with reactor model logs.

​
Updating a deployed model

Push a new version and publish it. Reactor performs a zero-downtime rollover: 
reactor model weights --model my-model --version v2 --source ./weights-v2/
docker build -t my-model:v2 .
docker tag my-model:v2 <your-registry>/my-model:v2
docker push <your-registry>/my-model:v2
reactor model publish --model my-model --version v2
The new version starts alongside the old one. Once healthy, traffic shifts over and the old version is drained. No downtime, no interrupted sessions.