User Management

Policy Management

Third Party Accounts

IAM Docs

iam_docs_image.svg

The vault to explore all the features  of Scalifi Ai’s 
Identity and Access Management.

Identity and Access Management

Documentation Hub

Model Designs

Model Jobs

Model Building Docs

The vault to explore all the features  of Scalifi Ai’s 
Model Building Service.

Model Building

Usage

Quota

BUC Docs

The vault to explore all the features  of Scalifi Ai’s 
Billing and Usage Center.

Billing and Usage

Model Management

Metadata

Model Version

Model Catalog Docs

The vault to explore all the features of Scalifi Ai’s Model Catalog.

Model Catalog

Model Versions vs. Model Variations

A model version represents a specific iteration or generation of a model. It is typically a snapshot of the model at a particular point in its lifecycle, often marked by significant changes or improvements made to the model's architecture, training data, or hyperparameters. Each model version encapsulates the complete setup needed to replicate the model's performance.

Example: Consider a facial recognition model. The initial release, trained on a limited dataset, is designated as Version 1.0. Over time, as the training dataset is expanded with more diverse images, the model is retrained to improve accuracy and robustness, resulting in Version 2.0.

Use Case: Model versions are used to track the evolution of a model over time. This allows developers and data scientists to deploy specific versions based on performance, compliance with regulations, or compatibility with specific applications.

A model variation refers to different configurations or customizations of a given model version within a specific framework. Variations might include changes in configuration settings, use of different layers or activation functions, or adjustments specific to deployment environments. Each variation within a version/framework combination is identified by a unique variation name.

Example: Continuing with the facial recognition model in Version 2.0, you might have different variations to address different operational needs:

Use Case: Model variations allow for fine-tuning and optimizing a model version to meet specific requirements or constraints of different deployment scenarios. They enable organizations to leverage the same underlying model architecture and training, while adapting to diverse needs and conditions.

Model Variation

A model version can have multiple frameworks, and each framework might consist of multiple variations with different configurations or customizations.

The platform currently facilitates a few frameworks, with additional ones currently in development and slated for imminent release. The frameworks list is:

Supported Frameworks

Add versions using client library

Add model versions

Add variations using client library

Add model variations

Under the Model Version section, expand the Model Variation Configuration section.

To add multiple tags, use the Add tag button. If any tag is not needed you can delete it by clicking on the delete button on the right side of the tag.

Finally, click on Submit to add the tags in a model.

Add model variation tags

Once the tags are removed from the model variation, you will no longer be able to categorize, search or filter that variation in relation to the tag key or values. Before deleting the tags make sure you no longer need those tags.

To remove model tags, simply click on the cross icon (X) of a tag.

To remove all model tags at once, click on the Remove all tags button.

Confirm the removal process and click on Confirm

Remove model variation tags

Refer the <a href='/docs/model-catalog/metadata'>metadata guide</a> for managing the model variation metadata.

Model variation metadata

Select the variant to delete from the dropdown. Then click on the Delete variant button.

Confirm the deletion process and click on Confirm.

Delete Model Variation

Select the version to delete from the dropdown. Then click on the Delete version button.

Confirm the deletion process and click on Confirm

Delete Model Version

Model versions refer to distinct iterations of a machine learning model, each capturing specific configurations and training states. This versioning system facilitates model tracking, comparison, and reproducibility, allowing data scientists to experiment, optimize, and collaborate effectively in the dynamic field of AI development.

Allowed operations that users can perform on the platform are:

But before moving on to the operations of model version lets see what is the difference between model version and model variation.