import-models-into-model-registry

ZMP Model Management provides two ways to register models into the Model Registry.

Getting Started

1. Select a Project: First, choose the project you want to work on.

2. Access the Model Registry Page: Click on "Model Registry" from the left menu or the homepage.

The Model Registry is managed on a per-project basis, allowing you to view all registered models at a glance.

Access New Model Screen

You can access the New Model screen from different entry points in the system:

Entry Point A: Start from Model List Screen

• From the System Admin Control Panel, click **Model **in the main navigation. Then, in the left-side navigation, select Model List.
• On the Model List screen, when the user clicks the Import from Hugging Face button or Create model repo button at the top-right corner, the system automatically navigates to the New Model screen.

Entry Point B: Start from Upload a model Screen

From the System Admin Control Panel, click **Model **in the main navigation. Then, in the left-side navigation, select Upload a model.

Importing a Model from Hugging Face

Hugging Face is a representative model hub where various open-source AI models are shared. You can directly select and register models that are publicly available on Hugging Face into the Model Registry without a separate download process. It allows for quick adoption and testing of external models, or efficient management when integrated with internal projects.

Step 1: Start Importing a Model

On the model creation type selection screen, select the **"From Hugging Face"** card.

The card displays the Hugging Face logo along with the description "Upload files via repo clone":

Step 2: Enter Basic Information

1. Migration Repo Name (Required)

• Enter a unique name for your model.
• This name will be used as the Git repository name.
• Example: my-bert-model,  gpt2-finetuned

2. Project (Displayed only if no project context exists)

• Choose the project to register the model under.

Step 3: Input Hugging Face Model Information

1. Hugging Face Model ID (Required)

• Enter the Hugging Face model ID.
• Example:
  • Model ID: google-bert/bert-base-uncased

2.** Click “Fetch Info” button**

• Click on the Fetch Info button to retrieve and preview the model information you entered.
• If you lack model access permissions, an error message will be displayed.

Step 4: Enter Authentication Details (Optional)

1. Username

• Enter your Hugging Face username.
• Needed for accessing private models.

2. Hugging Face Token

• Enter your Hugging Face Access Token.
• Required for private models or models that require authentication.

Step 5: Set Visibility

Default is private, you can choose one of the two options below:

• Private (project only): Accessible only within the selected project.
• Public: Accessible to all users.

Step 6: Complete Model Import

1. Once all information has been entered, click the "Save" button at the top right.

2. A success message will appear, and you will be automatically redirected to the Model List page.

3. Check that the imported model appears in the list.

4. Verify the model status to ensure the import was successful.

5. Model Statuses:

• READY: Migration completed successfully; model is ready for deployment.
• MIGRATING: Model migration is currently in progress.
• FAILED: Migration failed; click "detail" to view the cause of failure.

• Keep your Hugging Face Token secure.

• For private models, ensure you have the necessary access permissions.

• Importing large models may take some time.

Create Custom Model

Creating a private Model Repository enables you to register, store, and manage internally developed, customized, or self-created models. You can upload trained models and manage versions, centralizing model assets for easier collaboration and reuse.

Step 1: Start Creating a Model Repository

• On the model creation type selection screen, choose the Create model repo card.
• The card displays the Folder/File logo along with the description: "Upload files via repo clone":

Important Notice:

• For large model files (over 100MB), it is recommended to use Git LFS.

• Small configuration files can be uploaded directly via the browser.

Step 2: Enter Model Basic Information

1. Model Name (Required)

• Provide a unique name for your model.
• This name will be used as the Git repository name.
• Example: resnet50-custom, my-transformer-model

2. Project (Displayed only if no project context exists)

• Choose the project to register the model under.

3. Framework (Optional)

• Select the framework your model uses.
• Example: TensorFlow, PyTorch, ONNX, etc.

4. Library (Required)

• Select the library your model uses.
• Example: transformers, diffusers, timm, etc.

5. Task Type (Required)

• Select the task type for your model.
• Available options:
  • Text Classification
  • Token Classification
  • Question Answering
  • Translation
  • Summarization
  • Image Classification
  • Object Detection
  • Speech Recognition
  • Text to Speech

6. Version (Optional)

• Enter the version of your model.
• Example: 1.0.0, v2.1-beta

Step 3: Set Permissions and Metadata

1. Visibility

• Private (project only): Accessible only within the selected project (default)
• Public: Accessible to all users

2. Tags (Optional)

• Enter tags for easier model search, separated by commas.
• Example: custom, experimental, nlp

3. Description (Required)

• Provide a detailed description of your model.
• Suggested content:
  • Model purpose
  • Usage instructions
  • Training data
  • Performance metrics
  • Limitations

4. Model Icon (Optional)

   • Upload an icon to represent your model.
   • Supported formats: .png, .svg
   • Maximum file size: 5MB

Step 4: Complete Repository Creation

1. After entering all required information, click the Save button at the top right.

2. Upon success, you will be automatically redirected to the detailed page of the newly created model.

Step 5: Upload Model Files

Once the model repository is created, you can upload model files using the following methods:

1. In the Model Card, use the Clone Script to set up the repository in your local directory.

2. Upload large files using Git LFS (recommended for large files)

# Git LFS 초기화
git lfs install

# 대용량 파일 추적 설정
git lfs track "*.bin"
git lfs track "*.safetensors"
git lfs track "*.h5"

# 모델 파일 추가 및 커밋
git add .
git commit -m "Add model files"
git push

Example Model Repository Structure

A typical model repository might look like:

my-model/
├── config.json              # Model configuration
├── pytorch_model.bin        # Model weights (PyTorch)
├── tokenizer_config.json    # Tokenizer configuration
├── vocab.txt                # Vocabulary file
├── README.md                # Model documentation
└── .gitattributes           # Git LFS configuration