Hugging Face and Model Hubs
Level 4 · Where models live and run
Where open models are stored and shared. How to read a model page on Hugging Face, what Kaggle is for, and which uploads to trust.
Course contentsLesson 11 of 13
Level 1 · The AI labs
Level 2 · Names, sizes and licenses
Level 3 · Picture, video and speech models
Level 4 · Where models live and run
After this lesson you can
- Say what a model hub does and does not do
- Read a model card before you download
- Spot an upload you should not trust
10 mincovers 4 glossary entries
Worth reading first: Open Model Licenses
Where open models are stored and shared
A model hub is a website that stores AI models for people to find and download. Hugging Face, founded in 2016, is by far the largest: in November 2026 it listed over 3 million models. Each model has its own page with a model card. Kaggle, owned by Google since 2017, is a smaller hub built around contests.
Each model page holds the files, a model card and the license. The card says what the model is for, how it was made and where it fails. You can filter by task, size, language and license. Many pages let you try the model in the browser before you download it.
Hugging Face began as a chat app for teens, then turned into the place where AI is shared. It also makes free code that is widely used to load these models. Labs from Meta to DeepSeek post their open models there on the day they release them.
Kaggle began in April 2010 as a site for data science contests. Teams build models for a set problem and climb a leaderboard. It also hosts free data sets, and notebooks that run code in the browser. It has a library of models too. Many people learn by reading the notebooks that won past contests.
A hub hands you the files. It does not run them for you. Once you have the model, you need a machine with enough memory to run it. If you do not have one, an inference provider can run it for you and charge you for each use.
Anyone can post a model, so not every upload is safe or honest. Check who posted it, and whether it is the lab's own account. Read the license. Prefer files in the safetensors format, which cannot run hidden code when they are loaded.
People judge a model by its download count alone. A popular model may be old, or tuned for a task you do not have. Read the card, check the date, and test it on a few of your own examples before you build on it.
Worked example
Emma finds a model for receipts
Emma needs a model that reads Japanese receipts. She filters Hugging Face by task and language and finds 40 that might fit.
She keeps the 3 posted by known labs, reads their cards, and tests each on 20 of her own receipts. One gets 19 of 20 right.
The words for this lesson4
Check yourself
Does Hugging Face run the models you find there for you?
Finished this lesson?
Marking it complete also marks its 4 glossary entries learned, which updates your roadmap and your saved page. 0 are already marked.