Improving Hugging Face Model Access for Kaggle Users

Source: HuggingFace Date: 2025-05-14 URL: https://huggingface.co/blog/kaggle-integration

Summary

Integration announcement: HuggingFace and Kaggle launch bidirectional model access. HF model pages get a “Use this model” button that opens a pre-populated Kaggle notebook; Kaggle auto-generates model pages for any HF model referenced in public notebooks. Access-gated and private models require HF_TOKEN authentication and don’t generate Kaggle pages. Offline notebook submission for competitions (to prevent data contamination) is in progress but not yet available.

Implications

HF as open-source ML hub. Kaggle has ~10M data science users; embedding HF model discovery into Kaggle’s notebook workflow is a significant distribution channel expansion. Auto-generated Kaggle pages from notebook references means the HF catalog becomes visible to Kaggle users without any explicit HF team action per model.

Open-weights ecosystem health. The future offline submission feature — running notebooks without internet access during competition — is the harder and more important capability for open-weights adoption in structured ML competitions. Until that ships, HF models in Kaggle competitions require internet access during inference, which limits their use in prize competitions with strict evaluation environments.