Artificial Intelligence Startups Hugging Face recently announced that it is being led byThe two-person H4 team is working on developing tools and 'recipes' to help the AI community build ChatGPT-likeArtificial Intelligence ChatbotThe H4 team was formed to replicate the functionality of ChatGPT since its release. Since the release of ChatGPT, the H4 team was formed to replicate its functionality, using open source libraries and models.
H4 Team - name stands for 'useful, honest, harmless and friendly'-Now available in several open-source large-scale language models, including the recently launchedFine-tuning of the Mistral 7B model released by French AI startup MistralIn addition, H4 forked the Abu Dhabi Institute for Technological Innovation's Falcon-40B model, modifying it to be more conducive to responding to requests in natural language.
In order to train its model, theThe H4 team - like the rest of the Hugging Face research team - relies on a dedicated cluster of more than 1,000 Nvidia A100 GPUsThe team members, Lewis Tunstall and Ed Beeching, work remotely from Europe but are supported by several teams from within Hugging Face, including the model testing and evaluation team. Team members Lewis Tunstall and Ed Beeching work remotely from Europe, but are supported by a number of teams from within Hugging Face, including the model testing and evaluation team.
The H4 team emphasized thatThe small team size is an intentional choice as it allows them to be more agile and adapt to the changing research environment. In addition, they have collaborated externally with teams such as LMSYS and Llamaindex.
Recently, the H4 team has been working on different alignment techniques and building tools to test how effective the techniques proposed by the community and industry really are. This month, the team released a manual with all the source code and datasets used to build Zephyr, and H4 plans to update this manual with future AI model releases.
Faced with questions about whether there was pressure from Hugging Face executives to commercialize the work, Tunstall said that H4 does not profit directly from its tools. However, he acknowledged that the tools do support Hugging Face's enterprise-level service, the Expert Accelerator Program, which provides guidance from Hugging Face teams to build custom AI solutions .
Asked if H4 sees itself competing with other open-source AI initiatives like EleutherAI and LAION, Beeching said that's not H4's goal. Instead, he said thatThe intention is to 'empower' the open AI community by releasing training code and datasets for the H4 chat model..
Beeching adds.Our work would not be possible without the many contributions from the community."