Back to the feed
infra

Extending Inductor for the IBM Spyre Accelerator

PyTorch Blog1 min read

The PyTorch-based software stack for IBM's Spyre accelerator has transitioned its torch.compile backend from sendnn to inductor, an open-source backend. This change aims to leverage the capabilities of inductor for improved performance. The move is part of ongoing efforts to optimize the Spyre accelerator's software stack.

AI summary
Read at PyTorch Blog