Models and inference

Deploy a trained adapter

Turn a promoted adapter into an active inference endpoint, then create a token to call it.

You need the CLI signed in with the deploy scope, and a promoted artifact.

Find an artifact#

bash

gft deploy artifacts

bash

gft deploy start <promoted-artifact-id> --mode shared

start runs three steps in order: provision, canary test and activation. The deployment becomes active only if the canary passes.

Option	Meaning
`--mode`	`shared`, `dedicated` or `customer-account`.
`--account`	Provider account reference, kept on the server. Required for `customer-account`.
`--min-score`	Optional canary score floor. The benchmark score is the default.
`--yes`, `-y`	Skip the cost confirmation.

bash

gft deploy list
gft deploy show <deployment-id>
gft deploy token <deployment-id>
gft deploy remove <deployment-id>

gft deploy token creates a one-time inference credential. Provider credentials and endpoint URLs stay on the server.

Open Models to deploy an adapter, then try it in Playground.