Amazon’s TorchNeuron lets applications built with PyTorch run on its Trainium chips with a simple settings change.
The tool targets a key portability challenge in AI computing, where software support can tie applications to particular hardware.
Amazon says it plans to contribute code to PyTorch to help reduce that dependence.
How TorchNeuron works
Most developers today use PyTorch, a widely used tool for training AI models. Until recently, PyTorch's connection to Nvidia's software was so tight that Nvidia's components were the default choice for most development teams.
With TorchNeuron, Amazon allows PyTorch-based applications to run on its systems with a very simple change to their settings. This means that developers don't need to change the code of their programs to take advantage of Amazon's infrastructure.
The effort for greater independence
Amazon argues that using its own chips offers better price-performance ratio, while helping developers not be locked into one computing power supplier.
READ ALSO
Tolan app secures $30 million in new funding
In addition to the new tool, the company announced that it plans to contribute code to the open source PyTorch project — which has been owned by the Linux Foundation since 2022 — with the aim of reducing the software's dependency on specific chips. The goal is to make the software more flexible and work more easily on different systems.
It remains to be seen whether these changes will manage to change the market or whether developers will continue to prefer Nvidia's established solutions.
G
Hacks.gr on Google
Set it as a preferred source for cybersecurity updates.
Preferred source
