The two-person team at Hugging Face is developing ChatGPT-like AI models

Hugging Face

Startup Hugging Face has formed an H4 team of two developers to build chatbots based on open source AI models.

Here's What We Know

According to engineer Lewis Tunstall, after the release of ChatGPT, they started experimenting with using open source libraries and models to replicate such functionality. The main focus of H4 is training models to interact with humans.

The team has already presented several refined language models, including Zephyr-7B-α and Falcon-40B. A cluster of over 1000 Nvidia A100 graphics cards is being used to train them.

H4 is also investigating model alignment techniques and building tools to test them.

H4 deliberately stays small to be flexible and adapt to the rapidly changing AI landscape. In doing so, it collaborates with other Hugging Face research groups.

The team aims to empower the open source AI community by releasing code and data to train chat models. Commercialising their work is not a goal of H4, the developers noted.

Source: TechCrunch