Register and manage custom models

Custom models are used to register training artifacts, self-developed models, or external model files in Models. When registering a model, Specify its ownership, Version, storage location, tags, Description, and metadata. After registration, the model can be referenced in Develop, Training, or Inference.

​

Pre-Operation Confirmation

• The model files have been saved to storage accessible from the target region and cluster.
• The current account has permission to access the storage and path.
• The model name, version, and path have been confirmed to avoid difficulty distinguishing between model assets after registration.

​

Register a Custom Model

1. Go to Models > Custom.
2. Click Create.
3. Enter the model basic information.

   Parameter	Description
   Region and cluster	Determines the resource scope to which the model is registered and affects the scope of workloads that can subsequently reference the model.
   Name	A custom model name used to identify the model in the model list and when referencing the model.
   Type	The model task type.
   Version	The model version number, used to distinguish different iterations of the same model.
   Tags	Used to search for and categorize models.
   Description	Describes the model’s source, purpose, training data, or usage considerations.

4. Configure the model storage location.

   Parameter	Description
   Storage	Select the storage type where the model files are located. Currently, volume can be selected in the creation form.
   Storage	Select the storage resource for saving the model files.
   Dir	Enter the relative path of the model files in the storage.

   The Dir must be a relative path. Before registration, ensure that this directory contains the weights, configuration, tokenizer, and other files required to load the model later.
5. Enter metadata.

   Parameter	Description
   Architecture	Select or enter the model architecture to describe the model structure or series.
   Data Type	Select the data type used for model weights or inference.
   Parameters	Enter the number of model parameters, in B.
   Metrics	Add model evaluation metrics.

6. Click Create.

After submission, return to the custom model list. Search for models by region, cluster, or name, and verify that the model can be found and used in subsequent references.

​

Manage Custom Models

In Models > Custom, you can query custom models by region, cluster, and name. When managing models, pay particular attention to the model version, storage location, tags, and deployment status. To delete custom model records that are no longer needed, select one or more models and then delete them. Before deleting, ensure that the model is not depended on by any inference services, development environments, or training tasks that still need to be retained. The Deployed information in the list can be used to determine whether the model is in use.