Hugging Face & Google Cloud boost open AI model access

Hugging Face and Google Cloud have announced an expanded partnership designed to support enterprises in building their own artificial intelligence (AI) capabilities using open models. The collaboration aims to make open models more accessible and easier to deploy on Google Cloud's infrastructure.

Access to open models

Google Cloud customers have already been using open models from Hugging Face across a number of Google's AI platforms. Within Vertex AI, popular open models are accessible through Model Garden, while those seeking more control can use model libraries in GKE AI/ML or pre-configured environments managed by Hugging Face. AI inference workloads are being run on Cloud Run GPUs, supporting serverless deployments. These integrations enable businesses to choose how they deploy and manage their AI solutions across Google Cloud's services.

Expanded infrastructure

The partnership will introduce a Content Delivery Network (CDN) Gateway, optimised for hosting Hugging Face models and datasets on Google Cloud's infrastructure. This CDN Gateway, utilising Hugging Face's Xet storage technology and Google Cloud's networking resources, is expected to deliver faster model downloads and increased supply chain robustness. It will support platforms such as Vertex AI, GKE, Cloud Run, and Compute Engine.

According to the companies, usage of Hugging Face by Google Cloud customers has grown tenfold over the last three years. This growth now translates into tens of petabytes of model downloads every month. The new CDN Gateway is designed to respond to this scale, helping businesses reduce time-to-first-token and improve model governance.

Hugging Face tools

Hugging Face says its Inference Endpoints product will now be able to leverage Google Cloud's resources, including access to more instance types and potential price reductions. These enhancements are aimed at the large base of AI developers using the Hugging Face platform, supporting model deployment directly from Hugging Face to Google Cloud's environment.

With Google Cloud's custom AI accelerator chips, Tensor Processing Units (TPUs), now in their seventh generation, Hugging Face is working to provide native support for these chips within its libraries. The goal is to simplify the adoption of TPUs for running open models on Google Cloud, placing them on par with GPU compatibility for users of Hugging Face tools.

Security initiatives

The companies are also expanding security measures for open models hosted by Hugging Face. Hugging Face is set to use Google's security technologies, including VirusTotal, Google Threat Intelligence, and Mandiant, to help secure hosted models, datasets, and spaces. This focus aims to enhance supply chain security for developers and organisations making use of the Hugging Face Hub.

Industry outlook

"Google has made some of the most impactful contributions to open AI, from the OG transformer to the Gemma models. I believe in a future where all companies will build and customise their own AI. With this new strategic partnership, we're making it easy to do on Google Cloud," said Jeff Boudier, Product and Growth, Hugging Face.

Speaking for Google Cloud, Ryan J. Salva, Senior Director of Product Management, said, "Hugging Face has been the driving force enabling companies large and small all over the world to access, use and customise now more than 2 million open models, and we've been proud to contribute over 1,000 of our models to the community. Together we will make Google Cloud the best place to build with open models."