New Release: Ultralytics v8.4.163

Discussion
, ,
1

:rocket: Ultralytics v8.4.163: Export Apple Core AI Models from Linux

Quick overview: Ultralytics YOLO can now export Apple Core AI .aimodel files on x86_64 Linux, making cloud servers and CI workflows useful for preparing models for Apple devices. This release also improves export reliability and trims Docker images.

:red_apple: New Feature: Core AI Export on Linux

Export .aimodel files on x86_64 Linux or on Apple silicon Macs running macOS 26 or later. The exported models target iOS 27 and macOS 27; running them still requires Apple hardware. Core ML remains a separate export option. See the Linux Core AI export PR by @glenn-jocher.

:wrench: Improvements and Fixes

  • Updated compatibility: Support for coreai-torch 0.4.3 brings conversion-time optimization and Python 3.11–3.14 support. Compatibility PR by @cdeil.
  • Clearer benchmarks: Linux can report Core AI export results while correctly skipping inference, which requires macOS. Export and Docker PR by @glenn-jocher.
  • Safer failed exports: PyTorch functions patched for certain ONNX exports are restored even if export fails, avoiding unexpected effects later in the same process. Reliability PR by @Nikhi00718.
  • Leaner Docker images: Builds no longer retain package-install caches or bundled PyTorch test files, while keeping needed runtime tools. Export and Docker PR by @glenn-jocher.

Ready to give it a try? Explore the v8.4.163 release, review the full changelog, and let us know how Core AI export works in your workflow! :speech_balloon: