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

# Deploy a model

> Serve a fine-tuned model or any Hugging Face model behind an OpenAI-compatible endpoint.

A **deployment** is a running model you can call over an OpenAI-compatible API.
You can deploy a model you fine-tuned on Valkyrie, or any Hugging Face model
directly, no fine-tuning required.

## Deploy a base model

<Steps>
  <Step title="Open Models → Deploy">
    In the dashboard, start a new deployment.
  </Step>

  <Step title="Pick the model">
    Choose a base model (for example, a Qwen chat model) or paste a Hugging Face
    model id.
  </Step>

  <Step title="Launch and wait for Running">
    Valkyrie takes the deployment through a series of stages and it's only ready to
    use once it reaches **Running**:

    ```
    Queued → Provisioning → Downloading weights → Initializing → Health check → Running
    ```
  </Step>
</Steps>

## Deploy a fine-tuned model

From a completed [fine-tuning job](/guides/fine-tuning), deploy its output directly.
Valkyrie loads your trained weights or adapters onto the serving engine the same way
as a base model.

## Call your deployment

Once ready, the deployment exposes an OpenAI-compatible endpoint:

```
https://valkyrie-back.azumo.com/{deployment_slug}/v1/chat/completions
```

See [Run inference](/guides/inference) for full request examples, and
[Aliases](/guides/aliases) to give the deployment a stable, friendly URL.

## Managing a deployment

Deployments are designed to be operated over their whole lifecycle:

| Action            | What it does                                           |
| ----------------- | ------------------------------------------------------ |
| **Stop / Resume** | Release the GPU when idle, bring it back on demand     |
| **Restart**       | Restart the serving process without redeploying        |
| **Health check**  | Verify the model is actually serving requests          |
| **Schedules**     | Automatically start/stop on a cron schedule            |
| **Auto-restart**  | Recover the service automatically if it goes unhealthy |

<Tip>
  Idle deployments can keep a GPU running and cost money. Use **stop/resume** or a
  **schedule** so you only pay while the model is in use. See
  [Quota & billing](/guides/wallet-billing).
</Tip>

## Access and API keys

A deployment is called with an API key. Deployment API keys are prefixed `vk_dep_`.
You can share a model with teammates and manage who holds keys against it from the
model's detail screen. See [Models](/dashboard/models).