Amazon.com’s cloud unit on Wednesday said it has partnered with artificial intelligence startup Hugging Face to make it easier to run thousands of AI models on Amazon’s custom computing chips.
Valued at $4.5 billion, Hugging Face has become a central hub for AI researchers and developers to share chatbots and other AI software and is backed by Amazon, Alphabet’s Google and Nvidia, among others. It is the main place that developers go to obtain and tinker with open-source AI models such as Meta Platforms’ Llama 3.
But once developers have tweaked an open-source AI model they typically want to use the model to power a piece of software. On Wednesday, Amazon and Hugging Face said they had paired up to make it possible to do so on a custom Amazon Web Services (AWS) chip called Inferentia2.
“One thing that’s very important to us is efficiency – making sure that as many people as possible can run models and that they can run them in the most cost effective way,” said Jeff Boudier, head of product and growth at Hugging Face.
For its part, AWS is hoping to lure more AI developers to use its cloud services for delivering AI. While Nvidia dominates the market for training models, AWS argues it chips can then operate those trained models – a process called inference – at lower cost over time.
“You train these models maybe once a month. But you may be running inference against them tens of thousands of times an hour. That’s where Inferentia2 really shines,” said Matt Wood, who oversees artificial intelligence products at AWS.
