Hugging Face is a company and a platform where AI models, datasets and demos are published and found. Its main site, the Hub, works like a GitHub for AI. It also maintains open source libraries, such as transformers, to use those models from your code. It is not a specific model.
A question from the app. Pick an answer.
A tutorial says "get it from Hugging Face". What is Hugging Face?
That sounds like Ollama. Hugging Face is mainly where models are published and found.
It is a Hub, like a GitHub for AI. It also makes libraries, like transformers, to use those models from your code.
It is a platform, not a model: Llama is one of the many it hosts. Read the card (it may be incomplete) and the license: if there is none, do not assume you may use it.
Three main things. Models: files with the weights published by companies, universities and individuals, many of them open weights. Datasets: the data to train or test models. Spaces: demos you try in the browser, without installing anything.
When a tutorial says "get it from Hugging Face", it almost always means a model on the Hub. Then you run it with another tool, such as the transformers library, PyTorch or Ollama, if you want a local model.
transformers is a Python library to load a model from the Hub and use it with a few lines of code. It is named after the transformer architecture, but it works with many kinds of models. There are others, like datasets for handling data and diffusers for image generation models.
Look for the model card: what the model is for, how it was trained, its known limits and biases. Whoever publishes the model writes it, and it may be incomplete. Next to it you see the license. Some allow almost any use; others forbid commercial use or set conditions. If there is no license, do not assume you can use it. Some models ask you to accept terms before downloading.
"Hugging Face is a model, like Llama." No: it hosts a huge number of models, Llama among them. "If it is on Hugging Face, I can use it for anything." It depends on each model's license: read it before using it in a product. "Hugging Face is for running models on your computer." That is one thing you can do with what you download, not what it is.