In ContextSearch, the system provides built\-in inference services and also supports creating user\-defined inference services. When creating a user\-defined inference service, link it with the target model and configure network, resources, and user\-defined parameters based on business needs. This topic describes how to create a user\-defined inference service. ## Prerequisites Identify the model linked to the inference service and obtain the target model name in advance. * If linking a user\-defined model, create the user\-defined model first; see [Create a custom model](https://www.volcengine.com/docs/86958/1895487). * If linking a common model, obtain the common model’s name, dimension, and other details in advance; see [View a public model](https://www.volcengine.com/docs/86958/1903178). ## Procedure 1. Log on to the [ContextSearch console](https://console.volcengine.com/context-search). 2. In the top navigation bar, select the target project and region. 3. In the left navigation bar, select **Model and reasoning \> Inference service**, click the **Custom inference service** tab, then click **Create inference service**. 4. In the **Create inference service** panel, configure the following parameters, then click **OK**. |**Parameter** |**Description** | |---|---| |Affiliated project |Project of the inference service. Cannot be changed when creating an inference service. To modify it, close the panel and change it via the top navigation bar. | |Service name |Customize the name of the inference service.

* First character only supports letters or underscores (_).

* Can include letters, digits, and special characters: only English periods (.), underscores (_), hyphens (\-), and slashes (/) are supported, length between 1 and 128 characters.

* Can contain at most one slash (/) character. | |Description (optional) |Set the description for the inference service, with a maximum length of 256 characters. | |Tag (optional) |Add tags to inference services for easier identification and management. You can click **Add row** multiple times to add several tags. | |Select model |Select the public model or custom model associated with the inference service from the dropdown. Supports filtering by type or fuzzy search by model name keywords. | |Purchase method |Select **Exclusive resource** or **By token**. Some models do not support configuring this parameter. | |Resource type |Visible when **Purchase method** is set to **Exclusive resource** Select **CPU** or **GPU**. | |Specification |Visible when **Purchase method** is set to **Exclusive resource** Select CPU or GPU specifications from the dropdown list. | |Quantity |Visible when **Purchase method** is set to **Exclusive resource** Set the number of nodes for the inference service, ranging from 1 to 512. | |Private network |Visible when **Purchase method** is set to **Exclusive resource** Select the Virtual Private Cloud where the inference service resides. To create a new Virtual Private Cloud, see [Creating a VPC](https://www.volcengine.com/docs/6401/69467). | |Subnet |Visible when **Purchase method** is set to **Exclusive resource** The system automatically filters available subnets based on the Virtual Private Cloud you select. If you want to create a new subnet, see [Creating a subnet](https://www.volcengine.com/docs/6401/67823). | |Model rate limiting |Visible only when **Purchase method** is set to **By token**. Current account’s rate limit for accessing the selected model. | |Access point rate limiting |Turn on to set rate limiting for each access point. If off, the default is to limit by model.

* RPM: When access point rate limiting is on, you need to configure RPM (Requests Per Minute), which is the number of requests processed per minute.

* TPM: Visible when selecting the Volcano Ark model and the **Purchase method** is set to **By token**. When access point rate limiting is on, configure TPM (Tokens Per Minute), which is the number of tokens processed per minute. | |User\-defined parameter (optional) |Supports configuring user\-defined parameters for inference services. Click **Add custom parameters**, then configure the parameter name and value. |