Hugging Face raises $15 million to build open source community for cutting-edge conversational AI

Hugging Face has announced the close of a $15 million series A funding round led by Lux Capital, with participation from Salesforce chief scientist Richard Socher and OpenAI CTO Greg Brockman, as well as Betaworks and A.Capital.

New-York based Hugging Face started as a chatbot company, but then began to use Transformers, an approach to conversational AI that’s become a foundation for state-of-the-art algorithms. The startup expands access to conversational AI by creating abstraction layers for developers and manufacturers to quickly adopt cutting-edge conversational AI, like Google’s BERT and XLNet and OpenAI’s GPT-2 or AI for edge devices. More than 1,000 companies use Hugging Face solutions today, including Microsoft’s Bing.

The funding will be used to grow the Hugging Face team and continue development of an open source community for conversational AI. Efforts will include making it easier for contributors to add models to Hugging Face libraries and the release of additional open source tech, like a tokenizer.

Hugging Face CEO Clément Delangue said tech companies can’t build open source NLP on their own because research and engineering teams appear to be completely disconnected. Hugging Face and the 200 contributors to its open source project instead focus on providing state-of-the-art performance.

“In addition, the tech giants are not taking a truly open source approach, and their research and engineering teams are totally disconnected. On [the] one hand, they provide black-box NLP APIs — like Amazon Comprehend or Google APIs — that are neither state-of-the-art nor flexible enough. On the other hand, they release science open source repositories that are extremely hard to use and not maintained (BERT’s last release is from May and only counts 27 contributors),” Delangue told VentureBeat in an email.

Hugging Face libraries have thus far collected more than a million Pip installs from users of popular machine learning frameworks like TensorFlow and PyTorch. As part of its effort to make state-of-the-art NLP widely available, Hugging Face brought its popular library to TensorFlow in September.

Hugging Face launched in 2016 and is based in Paris and New York. The company has 15 employees and has raised $20 million to date.