Julien Chaumond, the technological director of one of the most popular startups, Hugging Face, lamented two weeks earlier about the lack of an alternative to OpenAI’s GPT Store. This story is no longer relevant following the launch of Hugging Chat Assistants.
The principle is simple. It consists of providing (free) users with the possibility of accessing a marketplace of artificial intelligence-based assistants – to exploit existing assistants or create new ones. And of course, these conversational robots must be designed using open source components accessible on Hugging Face.
Free and open source wizard portal
Since Hugging Chat Assistants went online on Friday February 2, Julien Chaumond has claimed “impressive” adoption. In the space of a few days, 4000 assistants were created – variously evolved.
This new service from the artificial intelligence startup is set to evolve. “Based on initial feedback from the community”, the CTO lists several features under study, such as the possibility of modifying its assistants via API.
RAG (retrieval augmented generation), at the heart of deployments and experiments in generative AI, and web search could also boost in the future the capabilities of assistants developed via open source bricks accessible on Hugging Face.
Users also welcome regular template updates for HuggingChat and Assistants. Because in addition to offering a marketplace of open source models and datasets, the startup provides tools developed using these components.
Alternatives to proprietary AI
This is the case of HuggingChat, a conversational assistant constituting an alternative to ChatGPT from the American giant OpenAI. To develop this robot, the Franco-American startup used the model designed by the French from Mistral AI, Mixtral-8x7B.
According to Julien Simon, the Chief Evangelist of Hugging Face, who spoke at the IMAgine Day IAGen on January 22, HuggingChat represents “a 100% open source alternative to proprietary chabots” and whose performance is there.
The young company founded by the French is developing other services, such as Enterprise Hub, and especially open source libraries. “The models alone are not of enormous interest. They are complicated to use,” explained Julien Simon.
Hugging Face allies of hyperscalers
The libraries thus simplify the use of these models, for example during the training and inference phases. With Spaces, the startup also provides application hosting on its own infrastructure or Inference Endpoints for deployment.
The strength of Hugging Face is also its integration with the main cloud platforms on the market. In January, its Chief Evangelist highlighted the strong links between the platform and that of AWS, the first hyperscaler to join forces with the startup.
Amazon’s advantage in terms of integration with Hugging Face may not be around for much longer. At the end of January, the AI gem signed a strategic partnership with Google Cloud. The agreement notably plans to make Vertex AI and GKE deployment options during the first half of 2024.