AWS launches TorchNeuron, letting PyTorch code run natively on its Trainium chips
AWS released TorchNeuron in public beta, an open-source native PyTorch backend for Trainium2 and Trainium3. Developers switch with .to('neuron'), and Deepgram reports about 60% more audio per dollar.
By the Bay Area Brief staff · Edited by Seamus McKenna · · 3 min read
Downtown San Jose and the Santa Clara Valley from the hills. Photo: Michael from San Jose, California, USA / Wikimedia Commons, CC BY 2.0 (cropped, resized and color-graded). Archive photo of the area; not a photo of the events in this story.
Amazon Web Services has released TorchNeuron, an open-source backend that lets PyTorch run natively on its Trainium AI chips. The public beta was announced Oct. 20 and is available under the Apache 2.0 license, according to the PyTorch blog.
Why it matters: PyTorch is the default tool for many AI researchers, and easier support for non-Nvidia chips could give AI teams cheaper options. It follows Google's similar TorchTPU move for its own chips.
How it works
AWS says developers can move PyTorch code to Trainium by switching to .to('neuron'), while keeping eager execution, torch.compile and standard PyTorch APIs. It also connects to the Neuron Kernel Interface for custom operators. It works on EC2 Trainium2 and Trainium3 instances.
Early users
The PyTorch blog says Deepgram reported about 60% more audio seconds processed per dollar than its GPU baselines. Hugging Face and Cartesia are also quoted in the announcement, which was unveiled at PyTorch Conference North America.
Sources
This article is an original summary based on the following reporting. We encourage readers to visit the original sources for full coverage.
The day's Bay Area housing, business and neighborhood news, plus the world headlines that matter, in one short email. Free, and you can unsubscribe anytime.
We use your email only to send the Brief. Details in our privacy policy.
At Oracle AI World on Monday, Oracle announced Fusion Companion and Fusion Agentic Home, which let employees state a goal and have AI agents carry out work inside Oracle's business software.
San Francisco's Phinity Labs says its agents beat experienced hardware engineers on key chip metrics and that it already has an eight-figure annual revenue run rate from frontier AI labs.
The San Francisco company told investors its September annualized revenue was near $50 billion. The bigger figure reported last month counted partners' gross sales, and chip and cloud stocks fell on the news.
