Skip to main content
A sampler checkpoint deploys to a Baseten inference deployment that serves the base model with your LoRA adapter loaded. Checkpoints outlive the run, so you can deploy one while the run is active or after it ends. The deployment bills for its GPU while it’s live. Deploying needs an hf_access_token in workspace secrets to download the base weights from Hugging Face. A run with base_checkpoint_auth deploys with those credentials. Only runs that start from Hugging Face weights can be deployed.

Deploy from the CLI

To deploy a checkpoint: Run baseten loops checkpoint deploy with the checkpoint’s globally unique id (the id field from listing checkpoints), by checkpoint name scoped to a run, or with the Truss CLI:
Both Baseten CLI flags skip the checkpoint picker, but the deploy still prompts for a model name, GPU type, GPU count, and the Hugging Face secret name (default hf_access_token). A successful deploy prints the IDs you need to call the model:
Output
The deployment downloads base weights and starts the inference server before it reaches ACTIVE. Use the Baseten CLI for interactive checkpoint deployment. To inspect the generated Truss config before creating an inference deployment, pass --dry-run:
The command still prompts for deployment settings, then prints the generated config without creating a deployment. Review the GPU and serving settings. Rerun without --dry-run when you’re ready to deploy. To customize the config, save the printed YAML as config.yaml in a Truss directory and edit it. Deploy with baseten model push, using the same CLI profile as the preview and the checkpoint’s team:
The generated config doesn’t store the team. Without --team, model push uses the organization’s default team, which can differ from the checkpoint’s team and cause deployment to fail. The Hugging Face secret must also be available in the checkpoint’s team.

Deploy from the REST API

To deploy a checkpoint: Call POST /v1/loops/checkpoints/deploy. Replace <id> with the checkpoint’s globally unique id, choose an instance type id from the instance-type catalog, and set the model name and Hugging Face secret:
Request
The response returns the model and deployment IDs:
Response

Deploy from the dashboard

Every Loop has a page in the dashboard at app.baseten.co/training/loop/<run_id>, reachable from the Training tab. Its Checkpoints table lists each saved checkpoint with its bt:// path, type, and size. Deploy one from that table; the flow matches deploying job checkpoints.

Call the deployed model

Call the deployment’s OpenAI-compatible chat completions route. The URL includes the deployment ID, and model is the checkpoint name (checkpoint_id), not its globally unique id:
Request
Output
For request options, streaming, and client libraries, see calling your model.

Delete the deployment

Delete it with DELETE /v1/models/{model_id} when you’re done:
Request
Output
Deleting the deployment doesn’t touch the checkpoint. You can redeploy it any time.

Next steps