Kernels

Add Metal (Apple Silicon) build variants

#6

Add pre-built Metal kernel variants for Apple Silicon Macs.

Build variants:

  • torch210-metal-aarch64-darwin
  • torch29-metal-aarch64-darwin

These enable GPU-accelerated rotary embedding on MPS (Metal Performance Shaders) backend, tested on M1/M2/M3/M4 with macOS 14+.

217/217 tests passing across all dtypes and configurations.

Source: https://github.com/robtaylor/rotary-embedding

Ready to merge
This branch is ready to get merged automatically.

Sign up or log in to comment