Skip to content

Create Model

Model Management is used to add user-defined models or fine-tuning task results, and to maintain their metadata, deployment templates, and weight mount information. It consolidates deployable assets in the platform, reducing repeated configuration and making it easier for teams to reuse and quickly deploy models later. The following sections describe the steps for creating metadata and managing templates and weight files.

Create Model Metadata

  1. Enter LLM Studio, expand Model Serving in the left navigation bar, and click Model Management.

  2. On the model management page, click Create Model.

    create-model

  3. After filling in the parameters on the creation page, click Confirm, and the model is created successfully.

    Parameter Constraints / Description Remarks
    Model Name Required
    Length: Up to 63 characters
    Characters: Only lowercase letters, numbers, hyphens (-), or dots (.) are allowed
    Rule: Must start and end with a letter or number
    Model Tags Required. Multiple selections are supported Used to identify the model capability type
    Model Icon Optional Supports local upload, or an image URL starting with https://; supports SVG, PNG, JPG, and JPEG formats, with a file size smaller than 40 KB. The recommended size is 100 x 100 pixels
    Model Description Optional Up to 800 characters each for Chinese and English

    create-model

Manage Deployment Templates

Deployment templates are used to preset resource configurations and runtime configurations for models. When deploying a custom model, you can directly use a template to reduce repeated input.

Create a Deployment Template

  1. In the model management list, click the target model name to enter the model details page.

  2. Select the Deployment Templates tab, and click Create in the upper right corner of the list.

    create-model

  3. Fill in Basic Information, Resource Configuration, and Runtime Configuration in sequence, and then click Confirm.

    Parameter Constraints / Description Remarks
    Template Name Required
    Length: Up to 63 characters
    Characters: Only lowercase letters, numbers, hyphens (-), or dots (.) are allowed
    Rule: Must start and end with a letter or number
    Description Optional Up to 800 characters
    CPU / Memory Resource configuration
    GPU Optional Includes the GPU type, number of physical cards, computing power, and video memory
    Runtime Framework Select the inference runtime Switching the framework automatically brings up the default startup command
    Startup Command Can be modified as needed
    Environment Variables Optional Configured as key-value pairs

    To use a custom inference runtime, complete the runtime configuration on the operations side first. For details, see Custom LLM Inference Runtime.

    create-model

Edit and Delete a Deployment Template

  1. In the Deployment Templates tab, click the ┇ menu to the right of the target template.

  2. Select Edit/Delete to modify or delete the template.

Manage Model Weight Files

Model weight files associate the weight directory on a cluster with the mount path inside the container for the current model. When deploying a custom model, the corresponding weight files can be found through this mount path.

Create a Model Weight File

Prerequisites

  • The target cluster must have Hydra Agent installed and file storage created.
  • The weight files must already be uploaded to the corresponding directory of the file storage, or synchronized to the cluster through remote file preheat.

Steps

  1. On the model details page, select the Model Weight Files tab, and click Create in the upper right corner of the list.

    create-model

  2. After filling in the parameter information, click Confirm.

    Parameter Constraints / Description Remarks
    Cluster Required Only clusters with Agent installed and file storage created can be selected. If no file storage has been created for the current cluster, click Create Now to go to the file storage creation page
    Version Identifier Required Used to distinguish between different weight versions under the same cluster
    Storage Type Required Currently only File Storage is supported
    Storage Directory Required Can be entered manually, or selected from the file storage by clicking Select Directory; you can also click Create Now to manage files
    Mount Path Required The mount path of the weights inside the container. The default example is /mnt/path

    create-model

Update and Delete a Model Weight File

  1. In the Model Weight Files tab, click the ┇ menu to the right of the target record.
  2. Select Edit/Delete to modify or delete the record.

Next Steps

Comments