Google Launches 'TorchTPU' to Run 'PyTorch' on Its Chips

Google announced the launch of 'TorchTPU', a specialized version of the open-source 'PyTorch' framework that allows existing applications to run on the company's TPU chips through configuration adjustments instead of rewriting code, according to SiliconANGLE.

Alongside this, the company launched a specialized agent package for TPU chips and an updated version of 'LiteRT', in a move that expands computing options for model developers.

The move comes after Amazon announced a similar version of 'PyTorch' called 'TorchNeuron', reflecting an effort by cloud computing providers to break developers' lock-in to Nvidia software.

This trend targets the software lock-in that ties many models to Nvidia's platform, as porting applications to alternative chips reduces rebuilding costs.

What Do These Terms Mean?

Open Source: Software whose code is available to everyone to inspect, modify, and use, often without strict commercial restrictions.

TPU Chips: Processing units specifically designed by Google to accelerate artificial intelligence computations instead of relying on GPUs.

PyTorch Framework: A set of software tools used by developers to build and train artificial intelligence models.

Software Lock-in: The reliance of applications on a specific platform such that porting them to another platform becomes costly and difficult.