Hugging Face sees small AI models advancing robotics

Hugging Face, the AI startup known for open-source innovations, is developing smaller language models aimed at enhancing robotics and on-device AI.

Speaking at Web Summit in Lisbon, Thomas Wolf, Co-Founder and Chief Science Officer of Hugging Face, emphasised that smaller models are key for real-time applications.

"We want to deploy models in robots that are smarter, so we can start having robots that are not only on assembly lines, but also in the wild," Wolf said, noting the importance of low latency.

“You cannot wait two seconds so that your robots understand what's happening, and the only way we can do that is through a small language model," he added.

Wolf highlighted that these smaller models can handle many tasks previously thought to require larger models and can be deployed directly on devices like laptops and smartphones.

"If you think about this kind of game changer, you can have them running on your laptop," he said.

Earlier this year, Hugging Face introduced its SmolLM, a small-scale language model that performs efficiently with fewer parameters.

Wolf explained that a one-billion parameter model could match the performance of larger, ten-billion parameter models from last year.

"You have a 10 times smaller model that can reach roughly similar performance," he pointed out.

According to Wolf, training these models on tailored datasets enhances their utility for specific tasks such as data processing and speech recognition.

Adaptations include embedding "very tiny, tiny neural nets" to further refine model specialisation, likened to "putting a hat for a specific task."