Hugging Face, AWS Partner to Help Devs ‘Jump Start’ AI Use
AWS is partnering with Hugging Face, the machine learning company, to make it easier to use and train machine learning and artificial intelligence models, the companies announced this week. As part of that partnership, AWS will offer JumpStart templates that will allow web, frontend and other software developers to deploy models quickly while accessing them through an API, said Jeff Boudier, head of product and growth at Hugging Face.
“We want to make those models really easy and cheaper to use through our collaboration with AWS, where most of the business community today is doing machine learning,” Boudier told The New Stack. “That means new and better experiences on Amazon SageMaker to make it super easy to run any of our models, and also new solutions to train them and run them cheaper using the hardware accelerators that were built from the ground up for AI that AWS offers.”
Optimizing for AI
SageMaker is AWS’ cloud machine learning platform. Launched in 2017, it allows developers to create, train and deploy machine learning models, as well as to deploy those models on embedded systems and edge devices. JumpStart SageMaker is AWS’ machine learning hub that provides hundreds of built-in algorithms, pre-trained models and solution templates to help developers quickly get started with machine learning.
The Hugging Face deep learning containers, as AWS calls the collection, is “packed with optimized transformers, datasets and tokenizers libraries to enable you to fine-tune and deploy generative AI applications at scale in hours instead of weeks – with minimal code changes,” AWS reported in a blog post.
As part of the collaboration, Hugging Face also will gain access to 1,000 GPUs available for a supercomputer cluster running SageMaker. The New York-based company and its open source partners will use that cluster to train new models, Boudier said.
“[It’s] very important for the community, for our mission to democratize good machine learning, that there are good open source alternatives to the the latest AI models that are available; and to do that, we need to be able to train models ourselves,” he said. “That’s one of the exciting parts of this new partnership. We now have over 1,000 GPUs that are available to us to build with the community the next generation of open source models.”
Developer Consideration: Targeted Models, Cost and Deployment
More targeted models may be the key to broader adoption of machine learning, according to Boudier.
“Right now, there’s lots of hype looking at the capabilities, the very general capabilities that large language models offer,” he said. “But the way that we can sustainably build machine learning into apps is one where we’re much more trying to find the right tool for the job. That will then enable scaling in a cost-effective way.”
He also said that this doesn’t change the Hugging Face experience — developers can still play with models in their sandbox and deploy wherever they want.
“What it changes, though, is if you’re building in AWS, then you have much better solutions to take advantage of any of the Hugging Face models and to run them and scale them more efficiently,” he aid.
Hugging Face hopes to extend its reach beyond data scientists and machine learning engineers to more software developers, including frontend and web developers. It’s possible to leverage these models with an API, and JumpStart makes that process simpler for developers, Boudier said.
“One of the main focuses of our collaboration with the AWS engineering teams is to make things easier” Boudier said. “SageMaker is great for the data scientists who want to really control what they’re doing and build their own thing, but if you just want to get something that works out of the box, there is the JumpStart solution to get you started.”
The cost of operation is also important for software developers and the partnership with AWS will make it possible to drive down the cost of running the models at scale, Boudier added. That includes using Inference Endpoints, a Hugging Face solution that allows developers to deploy machine learning models on a fully managed infrastructure.
“Where the usage of those models has been something that is the domain of scientists and machine learning engineers, we’re able to make it accessible to software developers via these new experiences,” Boudier said.
