Skip to main content

Create A Model

Create Model allows users to build a private AI model by customizing its basic settings, metadata, and access controls. On the Model Hub service screen, click the Create button to go to the creation page. file

Step 1: Fill in Basic Information​

file In the Basic Information section, complete the following fields:

  • Model Title : Enter the name of your model (required).
  • Description : Provide a short description of the model’s purpose.
  • Base Model : Select the foundational model (e.g., GPT-4, LLaMA).
  • Model Size (Parameters) : Specify the model’s parameter size (e.g., 7B, 13B).
  • Whitelist IPs : Add IP addresses that are allowed to access the model.
    • If no rule is set, all IPs are allowed to download the model. Max 10 IP addresses or CIDR ranges, separated by commas
    • IPv4 format:
      • x.x.x.x, where each x ∈ [0, 255]
      • Optional CIDR suffix: /y, where y ∈ [0, 32] ✅ Valid: 192.168.0.0 ❌ Invalid: 192.168.300.1
  • Avatar : Upload an image to represent your model (optional).
  • Overview : Write a detailed overview using rich text formatting (bold, italic, lists, etc.).

Step 2: Add Tags​

file Use the Tags section to assign relevant keywords to your model for easier categorization and searchability. Tag Categories You can assign tags under the following predefined categories:

  • Frameworks : Specify the development framework used (e.g., PyTorch, TensorFlow).
  • Architecture : Define the model architecture (e.g., Bert, DistillBert, EfficientNet).
  • Subject : Indicate the domain or field the model is related to (e.g., Healthcare, Finance, Education).
  • Task : Describe the model’s primary function (e.g., Text Classification, Image Classification, Embedding).
  • Others : Add any additional tags that don’t fit the above categories.

Step 3: Preview​

The Preview panel on the right will display a live preview of your model based on the information you’ve entered.

Step 4: Navigation​

  • Click Cancel to discard changes and exit.
  • Click Submit to finalize and create the model.