Kaggle and Hugging Face users are a part of one AI community. That’s why we’re excited to announce our plans to bring our platforms and communities closer to raised serve AI developers in every single place.
Starting today, Kaggle is launching an integration that enhances visibility and discoverability for Hugging Face models directly on Kaggle.
Learn how to start
You’ll be able to navigate from Hugging Face models to Kaggle and vice versa. Start by visiting a Hugging Face model page like Qwen/Qwen3-1.7B. To make use of it in a Kaggle Notebook, you’ll be able to click on “Use this model” and choose “Kaggle” to open up a Kaggle notebook with a pre-populated code snippet to load the model. You’ll be able to do the identical from a Hugging Face model page on Kaggle by clicking the “Code” button.
While you run a notebook on Kaggle that references a model hosted on Hugging Face Hub, we are going to routinely generate a Hugging Face model page if one doesn’t exist already. You don’t must make any special changes to your code. Moreover, whenever you make your notebook public, it’s going to routinely show on the “Code” tab of the Kaggle model page.
Discover Hugging Face models and explore all of the community examples in public notebooks in a single place on Kaggle at https://www.kaggle.com/models. As more Hugging Face models are used on Kaggle, the variety of models and associated code examples you’ll be able to probe for inspiration will grow.
When browsing Hugging Face models on Kaggle, we intend to make it easy so that you can navigate back to Hugging Face to explore additional details, metadata, community usage in Hugging Face Spaces, discussion, and more. Simply click “Open in Hugging Face” on the Kaggle model page.
How does this work with private and consent-gated Hugging Face models?
Should you use a personal Hugging Face model in your Kaggle notebook, authenticate via your Hugging Face account as normal (add your HF_TOKEN within the “Add-ons > Secrets” menu within the notebook editor). A Hugging Face model page won’t be generated on Kaggle.
If you ought to access a consent-gated model in your Kaggle notebook, you’ll must request access to it using a Hugging Face account and follow the prompts in your browser’s Hugging Face model page as normal. Hugging Face has documentation to guide you thru this process. Otherwise, the combination will work the identical as for non-gated models.
What’s next
We’re actively working on an answer to seamlessly use Hugging Face models in Kaggle competitions that require offline notebook submissions. While this may take a couple of more months to finish, we consider the wait can be price it.
You’ll be able to read Kaggle’s position on “AI Competitions because the gold standard for empirical rigor for GenAI evaluation” to know why it’s so vital for us to get this a part of the combination right! But tl;dr – Kaggle is very sensitive to data leakage and its impact on model contamination. Our goal is to design this integration to preserve the integrity of our competitions and their vital role within the industry, while enabling seamless access for Kaggle competitors to construct with one of the best models from Hugging Face!
We’d love to listen to your feedback within the meantime – share your thoughts and concepts here!
Comfortable Kaggling!
