New Release: Ultralytics v8.3.199

Discussion
, ,
1

Ultralytics v8.3.199 — faster imports, smoother exports, modern GPU Docker :high_voltage::spouting_whale:

Ultralytics v8.3.199 delivers quicker startup with lazy model loading, more stable export/runtime behavior, and refreshed GPU Docker docs. Expect faster imports, safer defaults, and clearer tooling across the board.

  • Version: v8.3.199
  • TL;DR: Faster import ultralytics, standardized export outputs, smarter TensorRT installs, safer torch.compile, better tuning plots, updated GPU docs, and stronger CI coverage.
  • Note: YOLO11 remains our latest stable and recommended model for all use cases.

Highlights

  • Faster imports via lazy model loading while preserving the public API
  • Consistent export outputs for quantized NMS, simplifying downstream integration
  • Automatic, CUDA-matching TensorRT installs on Linux
  • Safer torch.compile defaults to avoid CUDA Graphs pitfalls
  • Cleaner Tuner plots by excluding zero-fitness points by default
  • Modernized GPU Docker guidance using NVIDIA Container Toolkit
  • Re-enabled GPU export tests and benchmarks where GPUs are available

New Features

Improvements

Bug Fixes

Quick start and helpful snippets

  • Update to the latest version:
pip install -U ultralytics
  • Import remains the same, with YOLO11 as the recommended default:
from ultralytics import YOLO
model = YOLO("yolo11n.pt")
  • Show zero-fitness points in Tuner plots (previous behavior):
from ultralytics.utils.plotting import plot_tune_results
plot_tune_results("tune_results.csv", exclude_zero_fitness_points=False)
  • Run the GPU Docker image using NVIDIA Container Toolkit:
sudo docker run -it --ipc=host --runtime=nvidia --gpus all ultralytics/ultralytics:latest

What’s changed (PR roll-up)

You can explore the detailed release page in the announcement for Ultralytics v8.3.199 and review every commit in the full changelog from v8.3.198 to v8.3.199.

Try it and share feedback

Upgrade, run your workloads, and let us know how it feels—especially import times, export stability, and GPU Docker usability. Your feedback helps the community and the Ultralytics team keep improving.