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

# Prepare LLM Inference Config

> Before creating an LLM inference service, understand the relationship between LLM Engine Config, inference templates, and service creation configurations.

Before creating an LLM inference service, determine whether to reuse existing configurations. If suitable Engine configurations and inference templates already exist, you can select them directly when creating the service. If you need to establish standard configurations for your team, first maintain the corresponding configuration objects.

## Configure Object

| Object | Purpose | When to pay attention |
| - | - | - |
| **LLM Engine Config** | Defines the engine name, version, image, and role rules for different execute types. | Before using a custom engine, a new engine version, a different execute type, or PD-separated deployment. |
| **LLM Inference Template** | Combines the model, engine, execute type, probes, monitoring, and some advanced configurations for reuse when creating services. | When multiple services need to reuse the same configuration, or when the team wants to standardize service creation. |
| **LLM Inference Service** | Starts online service instances based on the model, engine, execute type, and resource configuration. | After confirming the model, resources, ports, gateway, and access method for this deployment. |

Not every service creation requires creating a new Engine configuration or inference template. If an existing configuration meets the deployment requirements, you can proceed directly to the service creation process.

## Configure Large Model Engine

1. Go to **Inference > LLM Engine Config**.
2. Select a **Region** and **Cluster**.
3. Click **Create**.
4. Enter the engine configuration JSON.
5. Click **Submit**.

Common fields are as follows:

| Field | Description |
| - | - |
| `engine` | Engine Name, for example, `vllm`, `sglang`, `dynamo`, or `custom`. |
| `version` | Engine version. |
| `image` | Image used to run the engine. |
| `routerImage` | Image used by the router role. |
| `executeTypeRoles` | Defines the roles that need to be displayed and configured for different execute types. |
| `showRoleRules` | Controls whether fields such as role parameters, cmd, Image, specifications, and replica count are displayed. |

After creation, you can edit, clone, or delete existing Engine configurations as needed. Before deleting a configuration, ensure that no templates or services depend on it.

## Prepare Inference Templates

If multiple services reuse the same model, engine, probe, monitoring, or advanced configuration, you can first create an LLM inference template. After the template is created, enable **Use Recommended Template** and select the template on the LLM inference service creation form.

For information about creating, reusing, and managing templates, see [Manage LLM Inference Templates](/siverse/inference/manage-llm-inference-templates.md).

## Next Steps

* [Manage LLM Inference Templates](/siverse/inference/manage-llm-inference-templates.md)
* [Create LLM Inference Services](/siverse/inference/create-llm-inference-services.md)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.