What is Hugging Face - Introduction and Best Ways To Get Started

If you are interested in AI, to either incorporate into a project or just want to keep with the latest AI capabilities, you have to learn how to use Hugging Face.

Hugging Face is used by all of the big tech companies like Google, Meta, and many more. It has become the leading platform of hosting and testing out AI models - currently it hosts over 270,000 models.

The models cover a variety of uses for example:

Hugging Face website chat testing shows chat bubbles from 2 people in conversation

Chat Model

Text to Image

a picture of a dog and a field showing a description of the dog

Text to Speech

Image to Text

a picture of jimmy fallon dancing and a black background picture with his pose outlined to the right of the image

Hugging Face has 3 main areas of focus

Models

Hugging Face's main offering. On the left, filter by the type of model you are looking for. And after that sort by Trending, Most liked, etc...

Once you've clicked on a model, there's general information regarding the model on the left and section for testing out the model directly on HuggingFace (circled in red).

Being able to test out the model from Hugging Face is an amazing feature because the alternative is downloading the model to your local machine (models require lots of storage capacity not to mention powerful processing)

A picture of a hugging face model with some images on the left and a few fields on the right and a graph of installations

From here you can also "🚀 Deploy" the model onto cloud provider server or host it on Hugging Face for free(this is slow but enough to do a little testing).

Many of the models have a "files" tab that provides download links for any

or that are used in image generation tasks. Below, we'll show you how to access the models with code.

Datasets

Datasets are used to train your own models. This page has similar filters as the models page. Use the filters on the left to narrow down to the data you want.

left has a bunch of categories while the right has a bunch of dataset blocks that can be selected

Spaces

Use spaces to display apps/services you've built with AI and also to discover projects that others have made.

an image of colorful blocks with descriptions of what the hugging face space (collaborative workspace) offers

Arrow 1, to view what models were used in the creation of the space. Arrow 2 leads to all the files for the space, so use this to find out how the space works under the hood.

an image of some fields and also some dropdown menus with 2 red arrows pointing at buttons files and one with a paperclip on it

Accessing the Models Programmatically

a screenshot of downloads last month (140, 769) and a chart showing the downloads over time

a screenshot of downloads last month (~42000) and a chart showing the downloads over time

LLM Models

Diffusion Models (Usually related to images)

Transformers/Diffusers allow you to programmatically access the models provided on Hugging Face. Click on the "Use in Transformers/Diffusers" button to get the model name.

Here's a straightforward guide on how to use transformers - with examples.

Note: this will download the model onto you local machine

Inference API

Another way to access the models is using Hugging Face's Inference API. This is a lighter weight option that allows you to test out different models and is offered for free by Hugging Face.

Here's a guide on how to use it: Inference API Guide.

For any other questions regarding Hugging Face, feel free to visit the FAQ page. There are some great Youtube videos covering Hugging Face, I recommend taking a look on the tube as well 👍