Alauda AI
English

Share Models

TOC

Introduction

While the AI platform doesn't have a UI for sharing models yet, you can still do it manually. Before we dive into the steps, let's cover some essential knowledge.

To publish a large language model (LLM) as an inference service, you need to add necessary model card metadata, such as pipeline_tag (called Task Type on the platform) and library_name. This information is typically stored as metadata in the README.md file at the model's root directory.

Alternatively, you can add this metadata after uploading your model to the AI platform's model repository. From the File Management interface, click the "Edit Metadata" button, select the parameters from the pop-up, and save.

Once a model has this essential metadata, it's ready for publishing. A key point to remember is that you cannot edit metadata for models that are already shared. This is an intentional platform restriction.

Also, if you're not using the UI to add metadata (for example, you manually created the README.md file), you'll need to add a few topics to the model's GitLab repository. This ensures the platform can correctly identify that the model is ready for inference.

WARNING

GitLab Admin privileges are needed to proceed. Please reach out to your administrator.

Steps

Adding Model Card Metadata: Two Options

You have two distinct methods for adding model card metadata. You only need to use one of these; you don't have to do both.

Option 1: Using the UI

This is the most straightforward method. After uploading your model to the AI platform's repository, follow these steps:

  1. Navigate to the File Management interface.

  2. Click the "Edit Metadata" button.

  3. In the pop-up window, set the Task Type based on your model and the desired service. For the Framework, if you don't have a special requirement, select transformers.

  4. Save your changes.

Option 2: Directly Editing the README.md File

If you prefer to manage metadata directly in your model's files, you can add a metadata block to the very top of the README.md file.

Add the following metadata block, setting the pipeline_tag based on your model and the library_name to transformers unless specified otherwise.

---
pipeline_tag: text-classification
library_name: transformers
language: []
datasets: []
metrics: []
---
INFO

Note: If you use this method, you must also add specific topics to the model's GitLab repository for the platform to recognize the metadata.

  1. First, go to <your model>'s project in GitLab. On the left-hand navigation bar, click on Settings, then select General

  2. In the "Project topics" section, add the following parameters:

    • For pipeline_tag, the topic value should be "<pipeline_tag value>" (e.g., text-classification).

    • For library_name, the topic value should be "aml_framework_<library_name value>" (e.g., aml_framework_transformers).

  3. Save Your Changes, be sure to scroll to the bottom of the page and click the "Save changes" button to apply the update.

Change Project visibility from Private to Public

  1. Navigate to Project Settings:

    Go to your GitLab project. In the left-hand sidebar, click Settings, then select General. This will take you to the project's general settings page.

  2. Expand Visibility Options:

    On this page, look for the "Visibility, project features, permissions" section and click the arrow to expand it.

  3. Change Visibility Level:

    Under the expanded options, you'll see a "Project visibility" dropdown menu. Click on it and choose Public from the list.

  4. Save Your Changes:

    After selecting Public, be sure to scroll to the bottom of the page and click the "Save changes" button to apply the update.

    CAUTION

    Making the model repository Public exposes all repository files (including code and potentially model weights) and associated model metadata to all users with access to your instance. Before changing visibility, ensure no sensitive data.