What Hugging Face is
Hugging Face is a company, and huggingface.co is its website, the Hub, where people share open AI models, datasets and demo apps. Everything on the Hub lives in a repository, a version-controlled folder with commits and branches, much like a GitHub repository. The Hub documentation puts the count at more than 2M models, 1.5M datasets and 1.5M Spaces, its name for hosted demo apps.
You use it in a browser, from Python with the huggingface_hub library, or from a terminal with its hf command. Browsing and downloading public models is free with an account or without one. A PRO plan raises your rate limits and adds monthly credits for hosted models, and some features, such as hosting a Gradio demo, need a paid plan.
What Hugging Face is good for
- Finding an open model for a job: pick a task such as text generation on the models page, then narrow by license.
- Checking whether you may use it: every model page has a model card, and its license badge names the license before you build on it.
- Downloading model files:
hf downloadfetches a whole repository or only the config and weights your code reads, as the download guide shows. - Running a model in Python: the Transformers
pipeline()call loads a model from the Hub and runs it in one line. - Calling models too big for your computer: Inference Providers lets one access token reach several hosted providers.
- Sharing a small demo: a Gradio app hosted on a Space gets a public web address.
- Publishing your own model: one command,
hf upload, sends a model folder to your own repository.
When to use Hugging Face, and when to pick another tool
Reach for Hugging Face when you want the files, the license or the code behind an open model. For a finished assistant to talk to, go straight to a chat product.
| Your job | Use Hugging Face? | Or try |
|---|---|---|
| Find an open model and check its license | Yes, that is what the Hub is for | Ollama's model library, if you only want models that run in Ollama |
| Chat with an assistant or draft a document | No, there is nothing to chat with | ChatGPT, Claude or Gemini |
| Run an open model on your own computer | It supplies the files, and another tool runs them | Ollama or LM Studio |
| Call an open model from code without a GPU of your own | Yes, through Inference Providers | Ollama, for models small enough to run at home |
| Put a small model demo on a web page | Yes, with a Gradio app on a Space | Not needed if the demo stays on your machine |
| Connect apps and automate steps with no code | No | n8n or Zapier |
How to use Hugging Face, step by step
These six steps need no account. They go from the models list to a gated model, with one screenshot of the site for each.
Open huggingface.co/models. Under Tasks, click Text Generation, then open Sort and pick Most downloads. The list narrows to text generation models, the most downloaded first.
Type
gpt2in the search box at the top. A menu of matches drops down withopenai-community/gpt2first. Click it.
The page that opens is the model card. The tags under the model name show its task, the libraries that load it and its license. Look for the License badge before you build on any model. This one says
mit.
Click Files and versions (just Files in a narrow window), next to Model card. This is the repository itself:
config.jsonis 665 bytes of settings, andmodel.safetensorsholds the weights at 548 MB.
Go back to Model card and click Use this model. Its menu offers Transformers under Libraries, notebooks on Google Colab and Kaggle, and local apps such as vLLM.
Some models ask for your agreement before they open their files. Open google/gemma-2-2b while signed out. A box says you must accept the conditions to reach the files, and it ends in Log in and Sign Up buttons.
Screenshots of huggingface.co, taken on 6 November 2026.
The tutorials at the top of this page show these steps on screen. Signing in, making an access token and downloading with the hf command are covered in lesson 1.
An example to try
These two lines install the library and preview what downloading GPT-2 would fetch. A dry run writes nothing, and a public model needs no token, so it is safe to try first. The download guide describes the flag.
pip install -U "huggingface_hub"
hf download openai-community/gpt2 --dry-run- The output lists the files in the repository and the bytes each would add. Nothing lands on your disk.
- In the docs' own example run, the repository held 26 files. Most of the size was the same model saved again for other tools, and your run may list other numbers.
- Add the file names you want, as in
hf download gpt2 config.json model.safetensors, to fetch only those.
Tips for better results
- Filter by task first and by license second, then open the model card. The model card docs explain the keys behind the badges.
- When a repository holds the same weights in several formats, download the
safetensorscopy. The Transformers docs call the older.binfiles insecure. - Use the CLI guide commands to clean up:
hf cache lslists what is stored andhf cache rmdeletes what you name. - Make one access token per machine or app, so revoking one leaves the rest working.
- Old tutorials say
huggingface-cli login. The v1.0 migration guide lists that name as removed, and the command is nowhf auth login.
Limits to know
- Signed-out requests get the lowest rate limits, counted per IP address in 5-minute windows. Past the limit the Hub answers with error 429, and a missing token is the usual cause.
- A gated model opens only after its authors approve a request, and they can withdraw access later. Note an open alternative before you build on one.
- A license badge names the license and cannot say whether your use fits. Read the terms before a model goes into a product you sell.
- Hosting a Gradio demo on a Space is not free for everyone. The Spaces lesson shows which plan it needs.
Learn more about Hugging Face
- 1Hugging Face tutorial: find, check and download open modelsA Hugging Face tutorial for Python users: find a model on the Hub, read its card and license, request a gated model and download it with the hf CLI.
- 2Transformers pipeline: run open models in Python, on CPU or GPUA Transformers pipeline tutorial: install it, classify and generate text, chat with an instruction model, and choose a device and dtype that fit your memory.
- 3Hugging Face Inference API: call hosted open models with Inference ProvidersCall hosted open models through Inference Providers, the successor to the Hugging Face Inference API: a token, InferenceClient, the OpenAI client and curl.
- 4Hugging Face Spaces and Gradio tutorial: build a demo and host itA Gradio tutorial for Hugging Face Spaces: build a demo with gr.Interface, run it locally, then host it on a Space with its requirements, secrets and ZeroGPU.
Frequently asked questions
Is Hugging Face free?
Yes for the basics. An account, browsing and downloading public models cost nothing. A PRO plan raises your rate limits and adds monthly credits for hosted models, while free accounts get no monthly credits but can buy some.
Do I need an account to download a model?
Not for a public model. Gated models need an account and the authors' approval, and signed-out downloads get the lowest rate limits. A free account with an access token lifts those limits and is the better default for any script.
What is a model card?
It is the README of a model repository, shown at the top of the model page. A block of metadata at its start feeds the license badge, the task tag and the filters, and the text below says what the model is for and how to use it.
What is a gated model?
A model whose authors want a request before the files open. You agree to their conditions while signed in, and with automatic approval the files open at once. With manual approval you wait, and the authors can block your access later.
How do I download a model from Hugging Face?
Click Files and versions on the model page to download single files in the browser, or run hf download with the repository name. Add file names to fetch only those, and --dry-run to preview first.
Can I use a Hugging Face model in a product I sell?
It depends on the license on its card. Licenses such as apache-2.0 and mit are permissive, cc-by-nc-4.0 forbids commercial use and license: other points to custom terms you must read.
What is the difference between Hugging Face and Ollama?
Hugging Face hosts model files and the code to load them. Ollama downloads models and runs them on your computer, and it can pull GGUF repositories from the Hub with ollama run hf.co/{username}/{repository}.
Sources
- Hugging Face Hub documentation https://huggingface.co/docs/hub/index
- Hugging Face Hub: Model cards https://huggingface.co/docs/hub/model-cards
- Hugging Face Hub: Licenses https://huggingface.co/docs/hub/repositories-licenses
- Hugging Face Hub: Gated models https://huggingface.co/docs/hub/models-gated
- Hugging Face Hub: User access tokens https://huggingface.co/docs/hub/security-tokens
- Hugging Face Hub: Rate limits https://huggingface.co/docs/hub/rate-limits
- huggingface_hub: Command Line Interface (CLI) https://huggingface.co/docs/huggingface_hub/guides/cli
- Hugging Face: Download files from the Hub https://huggingface.co/docs/huggingface_hub/guides/download
- huggingface_hub: Migrating to v1.0 https://huggingface.co/docs/huggingface_hub/concepts/migration
- Hugging Face Hub: Spaces overview https://huggingface.co/docs/hub/spaces-overview
- Inference Providers overview https://huggingface.co/docs/inference-providers/index
- Transformers: Pipeline https://huggingface.co/docs/transformers/pipeline_tutorial