Ultralytics YOLO v8.2.90 Release Announcement
We’re excited to announce the release of Ultralytics YOLO v8.2.90, packed with enhancements to improve your experience and model performance. This update focuses on optimizing memory management for macOS users, updating dependencies for better stability, and refining user interactions.
Key Changes
Apple MPS Memory Optimization
- Improvement: Integrated
torch.mps.empty_cache()for macOS to enhance memory management when using MPS devices. - Impact: Expect up to a 40% reduction in training time, significantly boosting performance for macOS users.
ONNXSlim Dependency Update
- Change: Reverted
onnxslimto version0.1.32. - Reason: This update addresses export issues with YOLOv10 for TFLite, ensuring smoother model export processes.
Default Save Behavior
- Adjustment: The
saveoption now defaults toTruefor CLI andFalsefor Python scripts. - Benefit: Aligns with user expectations, providing automatic result saving in CLI and flexibility in Python environments.
Purpose & Impact
This release is designed to enhance resource management on macOS, ensure compatibility with updated dependencies, and improve user experience through intuitive interface adjustments. 
What’s Changed
- Revert to ONNXSlim 0.1.32 by @glenn-jocher in PR #16084
- MPS Unified Memory Cache Empty by @Oil3 in PR #16078
- Fix Visualization Arguments Docs Table by @MatthewNoyce in PR #16061
- Apple MPS Unified Memory Clearing by @glenn-jocher in PR #16085
Full Changelog: v8.2.89…v8.2.90
Try It Out!
We encourage you to explore the new features and improvements in this release. Your feedback is invaluable to us, so please share your experiences and suggestions.
For more details, visit the release page.
Happy experimenting!