Google LLC’s cloud unit today announced a new partnership with Hugging Face Inc., the operator of a popular platform for sharing open-source artificial intelligence models.

The deal will see Google Cloud become a “preferred destination” for Hugging Face’s AI training and inference workloads. Additionally, the companies intend to integrate several of their respective services with one another.

Hugging Face, which received a $4.5 billion valuation last August, provides a GitHub-like platform for hosting open-source AI models. The platform also lends itself to sharing related files such as training datasets. In addition to hosting other developers’ machine learning projects, the company has built two open-source language models through research partnerships.

The company monetizes its platform with a set of paid cloud services. One service, AutoTrain, promises to reduce the amount of manual work involved in training AI models. Hugging Face also offers Inference Endpoints, which enterprises can use to host AI models that have already been trained.

As part of its new partnership with Google Cloud, the company will roll out an integration for the search giant’s Vertex AI product suite. The suite includes AI development tools and more than 130 prepackaged foundation models. According to the companies, developers will gain the ability to access Vertex AI from the Hugging Face interface with a few clicks.

A similar integration will roll out for Google Kubernetes Engine, the search giant’s managed Kubernetes service. Developers can use the service to run AI workloads, such as open-source neural networks downloaded from Hugging Face, in the form of software containers. GKE makes it possible to run those containers on the AI-optimized chips that Google provides through its cloud platform.

The search giant says that Hugging Face users will have access to the Cloud TPU v5e, the latest iteration of its internally developed machine learning processor. The chip promises to provide up to 250% better performance per dollar than Google’s previous-generation silicon. It can also run AI models with less latency. 

Down the road, Hugging Face will add support for Google’s recently launched A3 VMs. Each such instance features eight H100 graphics cards from Nvidia Corp. that can exchange data with one another significantly faster than the chips in previous-generation instances, which can help improve AI performance. The GPUs in the A3 VMs are supported by two Intel Corp. processors and two terabytes of memory.

“Google Cloud and Hugging Face share a vision for making generative AI more accessible and impactful for developers,” said Google Cloud Chief Executive Thomas Kurian. “This partnership ensures that developers on Hugging Face will have access to Google Cloud’s purpose-built AI platform, Vertex AI, along with our secure infrastructure, which can accelerate the next generation of AI services and applications.”

Alongside the new infrastructure options, Hugging Face will add an integration with Google Cloud Marketplace. It’s designed to ease billing-related tasks for customers of Hugging Face’s paid AI development and hosting services. The integration will also work with the company’s Spaces portal, which open-source developers use to share AI applications. 

Photo: Google