ultralytics 8.3.244 on Python PyPI

🌟 Summary (single-line synopsis)

YOLO inference is now safer when switching devices mid-run by automatically resetting the predictor to avoid stale state issues 🔄🖥️✅

Predictor reset on device change (main update, PR #23086 by @Y-T-G) 🔁
- In ultralytics/engine/model.py, the model now detects when device= changes between predict() calls (e.g., CPU ↔ GPU) and re-initializes self.predictor instead of reusing the old one.
ExecuTorch export dependency fix for ARM64 Docker (PR #23064) 🐳🧩
- Removes the broad setuptools<71.0.0 workaround and replaces it with a targeted fix: on Linux + ARM64 + Docker, require packaging>=22.0 to avoid build/export issues.
Docker + examples polish and reliability improvements (PR #23090, #23092) 🐳🛠️📚
- Cleaner, more reproducible Docker builds (less layer bloat, more consistent installs).
- ONNXRuntime C++ demo preprocessing fixes (letterbox/shape handling), plus small hardening/robustness tweaks in examples and tests.
- Documentation/config wording cleanup (including references updated to YOLO11) ✍️

More reliable predictions in real apps/services 🚀
- If you run a long-lived process and sometimes call model.predict(device="cpu") and other times device=0, the predictor will no longer “carry over” device-specific state that can cause incorrect or surprising results.
Minimal impact for typical users 🙂
- If you keep the same device, behavior is effectively unchanged (it continues to reuse the existing predictor for efficiency).
Smoother exports and deployments 📦
- ARM64 Docker users exporting to ExecuTorch should see fewer dependency-related failures.
- Docker images and example integrations become easier to build/run consistently across environments.

refactor: ♻️ remove setuptools check_requirements in executorch export by @onuralpszr in #23064
Dockerfile improvements by @glenn-jocher in #23090
YAML fixes by @glenn-jocher in #23092
ultralytics 8.3.244 Reset YOLO predictor on device change by @Y-T-G in #23086

Full Changelog: v8.3.243...v8.3.244