Kaggle is launching Kaggle Models in a move that will allow community members to publish models and share them. The effort also puts Kaggle in competition with Hugging Face to some degree.
In a post, Kaggle outlined Kaggle Models, which is designed to be an open marketplace for stress testing machine learning and generative AI use cases.
Kaggle users, also known as Kagglers, can publish models via the "+ Create button" on the left of the home page. Kaggle has published a set of API publishing instructions. Kaggle models can be published in any framework and upload to 100GB per model variation.
Hugging Face developers to plug into Google Cloud infrastructure
Users will have to document the model variations including framework, model weights and license as well as document them. Models can be private to the user or made public.
Kaggle said best practices include the open-source approved license such as Apache 2.0, model cards for transparency and instructions with examples.
Going forward, Kaggle said it will listen to feedback and iterate, make it easier to publish models without a review process and publish directly. There will also be additions for organizations to share models and datasets.
