Hugging Face is an American developer of tools that enable users to build applications with machine learning.
The company, which was founded in 2016 by French entrepreneurs Julien Chaumond, Thomas Wolf, and Clément Delangue, initially offered an app-based chatbot for the teenage demographic.
Related Articles
A short history of Hugging Face
The chatbot was designed as a fun and entertaining communicator that could talk about a diverse range of subjects but also detect emotions in the user’s text and alter its tone to suit.
Underpinning this chatbot was an in-house natural language processing (NLP) Model and also a library of pre-trained models. These would serve as the foundation for early PyTorch transformers which facilitated compatibility between PyTorch and TensorFlow 2.0.
When the seminal paper Attention Is All You Need was released in 2017, the use of transformers shifted to the NLP space. The company, which had previously released part of the library powering its chatbot on GitHub, then standardized the differences between models that were trained with different libraries.
What does Hugging Face offer?
Open-source platform
The company’s open-source platform, known as HF Hub, provides free access to anyone who wants to explore, experiment, collaborate, and build tech with ML. HF Hub offers over 60,000 models, 6,000 datasets, and 6,000 ML demo apps.
Inside HF Hub, users will be able to create or access these ML assets:
- Models – initially, the hub was a repository for state-of-the-art NLP models but has since expanded to computer vision, speech, RL, chemistry, biology, and time-series models. Hugging Face incorporates transfer learning so that any team can use a pre-trained model and easily fine-tune it for their own purposes.
- Datasets – HF Hub datasets feature a variety of data for various modalities, languages, and domains. Users can locate the best dataset for their use case by using the search bar, and several tools make data manipulation fast and efficient.
- Spaces – these are interactive, browser-based apps that enable users to showcase their models to business stakeholders in particular.
The hub is also a space where individuals can solicit feedback on their models. Teams can also peer review each of the ML assets and create organizations where said assets can be managed collaboratively.
Private Hub
The Hugging Face Private Hub is a place where companies can utilize the company’s complete ecosystem in a private and compliant environment.
Private Hub was established in response to the fact that 87% of data science projects never make it into production, with the primary causes being siloed organizations, task and effort duplication, and data that lives in different formats.
To accelerate the ML development process and make collaboration simpler and more productive, Private Hub offers a unified set of tools from research to production.
Data scientists can now work with over 25 open-source transformer libraries, with over 10,000 companies such as AWS, Intel, Microsoft, Meta AI, and Grammarly using the technology.
Subscription-based NLP features for deployment
While the HF Hub and its various features and tools are free, the company does offer certain upgrades and additional features for a price.
Let’s take a look at a few of these below:
- Spaces Hardware – for those who want to upgrade their Spaces with custom hardware, various fees are charged. A basic CPU upgrade, for example, is a mere $0.03 per hour, while an upgrade to the Nvidia A100 GPU is $4.13 per hour.
- Inference Endpoints – a secure production solution that enables users to easily deploy any model on dedicated, autoscaling infrastructure. Various options exist. For Azure CPU instances on the Intel Xeon architecture, 8 vCPUs, and 16GB memory, Hugging Face charges $0.48 per hour.
- Pro Account – for $9 per month, users with a Pro account receive a special badge, unlimited models and datasets, early access to new features, and higher limits for both the AutoTrain and Free Inference API features.
- AutoTrain – a feature that creates, trains, evaluates and deploys state-of-the-art ML models without code. Free to use initially, but extra images, rows, and models are available for a price.
- Enterprise – for enterprise customers who want to train their own LLMs, Hugging Face sells support from over 40 machine learning experts. Prices are available on request.
Key takeaways:
- Hugging Face is a developer of tools that enable users to build applications with machine learning. The company, which was founded in 2016 by Julien Chaumond, Thomas Wolf, and Clément Delangue, was conceived initially to create an app-based chatbot for teenagers.
- The company’s open-source platform (HF Hub) provides free access to anyone who wants to explore, experiment, collaborate, and build tech with ML. HF Hub offers over 60,000 models, 6,000 datasets, and 6,000 ML demo apps. Private Hub is for clients who want to utilize the company’s complete ecosystem in a private and compliant environment.
- While the HF Hub and its various features and tools are free, the company does offer certain upgrades and additional features for a price. These include upgrades to Spaces, Inference Endpoints, and AutoTrain, as well as a premium support option for enterprises and a Pro account with access to various perks.
Read Next: History of OpenAI, AI Business Models, AI Economy.
Connected Business Model Analyses
AI Paradigm
Pre-Training
Large Language Models
Generative Models
Prompt Engineering
OpenAI Organizational Structure
OpenAI Business Model
OpenAI/Microsoft
Stability AI Business Model
Stability AI Ecosystem
The post What is Hugging Face? appeared first on FourWeekMBA.
