> ## Documentation Index
> Fetch the complete documentation index at: https://docs.primeintellect.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Deploy a training checkpoint

> Serve a saved training checkpoint through an OpenAI-compatible endpoint without a manual weight export.

Deploy a saved training checkpoint directly from a Model Factory run. Prime loads the checkpoint into a serving engine, so you do not need to download it or convert it to Hugging Face weights first.

## Check the prerequisites

* The run belongs to a team you can access.
* The checkpoint is saved and available as `READY`. Only complete checkpoints on a finalized volume can be served, so the run must be completed, stopped, or paused, and its final volume sync must have finished. Checkpoints still on an actively training run's volume are not available. [Pause](https://docs.primeintellect.ai/hosted-training/pause-resume.md) a running run to sync its volume and release its GPUs without waiting for it to finish.
* The run's base model has a serving configuration in the inference catalog. If deployment reports that the base model is missing, ask [support](mailto:contact@primeintellect.ai) to add it.
* On-demand deployment needs a funded team wallet and available capacity. Using the run's own cluster needs an admin of that cluster's team, with inference and self-service actions enabled.

For CLI access, install Prime CLI 0.21.0 or later. The key needs `factory:read` and `factory:write` for deployment management, plus **Inference** permission to query the endpoint.

## Deploy in the dashboard

1. Select the run's team, then open [Dedicated Endpoints](https://app.primeintellect.ai/dashboard/inference/dedicated).
2. Click **Create endpoint**, and select **Custom weights**.
3. Search for the training run, then choose one of its saved checkpoint steps.
4. Under **Deploy to**, choose the cluster the run trained on or **On-demand capacity** when both are offered. Review the cost before using on-demand GPUs.
5. Click **Create endpoint**, then wait for it to be running.

Custom weights are available in team accounts. The picker lists completed, stopped, and paused runs whose volume sync has finished — on-demand runs, and runs on a team cluster with inference enabled. A listed run with no complete checkpoint on its volume shows an empty checkpoint list.

## Deploy with the CLI

List ready checkpoints for your run:

```bash theme={null}
prime train checkpoints <run-id> --status READY
```

Choose a step from the output and pass it explicitly:

```bash theme={null}
prime inference deploy-checkpoint <run-id> --step 100 --wait
```

Replace `100` with an available step. The command has no default checkpoint step. The equivalent training command is:

```bash theme={null}
prime train deploy <run-id> --step 100 --wait
```

The CLI defaults to the run's team. For a run on a team-owned cluster, it deploys to that cluster. Otherwise, it uses on-demand capacity billed to the run's team. The CLI checkpoint command does not expose placement or initial replica options. Use the dashboard to choose on-demand capacity for a run that trained on a team cluster, or use the [API's `onDemand` field](https://docs.primeintellect.ai/api-reference/inference-deployments#deploy-a-checkpoint).

If that checkpoint already has a pending, deploying, or running deployment, the request returns the existing deployment, even when you ask for a different placement or replica count. Deploying another step creates a separate deployment. Stop the old one when you finish comparing them.

## Query and manage the endpoint

Inspect the deployment returned by the command:

```bash theme={null}
prime inference deployments get <deployment-id> -o json
```

Once its status is `RUNNING`, use `servedModelAlias` as your request's model identifier and `endpointUrl` as the OpenAI-compatible base URL. Use the run's team context for requests. See [Send a request](https://docs.primeintellect.ai/inference/dedicated-deployments.md#send-a-request) for CLI and Python examples.

The endpoint has its own lifecycle. Exiting the CLI does not stop it. Manage replicas and release the GPUs with:

```bash theme={null}
prime inference deployments scale <deployment-id> --replicas 2
prime inference deployments delete <deployment-id>
```

Keep the source checkpoint available while the endpoint is deployed. Workers that start or restart read it again, and a deployment whose checkpoint is gone fails with `The checkpoint is no longer available.` Deleting the inference deployment leaves the training run and source checkpoint intact.

If no checkpoints appear, see [Checkpoint unavailable](https://docs.primeintellect.ai/inference/troubleshooting.md#checkpoint-unavailable). To export weights for use outside Prime, follow [Retrieve and export the checkpoint](https://docs.primeintellect.ai/hosted-training/end-to-end-run.md#retrieve-the-checkpoint).
