Ultralytics v8.3.243 is out! 
TL;DR: v8.3.243 sharpens training telemetry and console output—cleaner progress bars, richer run metadata for Ultralytics Platform-style tracking, and safer/less noisy logging 
If you’re training Ultralytics YOLO models like YOLO11 (recommended), this release is all about making your runs easier to monitor, compare, and debug—especially in notebooks, CI, Docker, and streamed logs.
New Features (highlights first)
Cleaner ConsoleLogger progress bars (dedup + batching)
Training output is now significantly less spammy:
- Suppresses noisy in-progress bar updates (while keeping completion signals)
- Prevents repeated
100%lines for the same step (epoch/train/val/Class) - Adds batched flushing (by lines/time)
- Safer handler cleanup
Merged in PR #23082 by @glenn-jocher and PR #23079 by @glenn-jocher
Richer training_started metadata for telemetry
Run tracking gets a big boost with a new _get_environment_info() collector, adding:
- OS, Python, hostname
- CPU/GPU details
- command used to launch training
- Git repo/branch/commit (when available)
- model stats (params, GFLOPs, class count)
Included in PR #23079 by @glenn-jocher
Checkpoint upload timing tweak (rate-limit initialization)
The upload timer now initializes at training start, so the first checkpoint upload waits ~15 minutes—reducing early-run upload churn.
Included in PR #23079 by @glenn-jocher
Docs & Usability Improvements
Headless install guidance promoted (servers/CI/Docker)
Docs now recommend ultralytics-opencv-headless for headless environments to avoid OpenCV GUI libGL issues.
Shipped in PR #23077 by @glenn-jocher
Minimal example:
pip install ultralytics-opencv-headless
Docker GPU run instructions clarified
Docker examples now include --runtime=nvidia to reduce “GPU not found” surprises.
Updated in PR #23080 by @lakshanthad
Opt-out of auto requirements checks
New environment variable to skip requirements checks/auto-install behavior (useful for locked/controlled environments):
Added in PR #23065 by @onuralpszr
export ULTRALYTICS_SKIP_REQUIREMENTS_CHECKS=1
Small Fixes / Notes
- Pose loss comments corrected in PR #23060 by @raimbekovm
- YOLO-World docs note a community tip (adding
""as background class may help in some cases) in PR #23058 by @raimbekovm - Dependency update (
zensical>=0.0.15) in PR #23052 by @glenn-jocher
Why this release matters
- Cleaner logs : easier to spot real warnings/metrics in notebooks, CI, and forwarded logs
- Better reproducibility
: environment + Git + model metadata improves traceability across machines/containers/branches - More predictable syncing : reduces early bandwidth spikes from checkpoint uploads
- Smoother installs
: fewer dependency headaches on servers and pipelines - No major model behavior changes : focused on logging/telemetry/usability/docs
Upgrade & try it
Install or upgrade:
pip install -U ultralytics
Then run a quick sanity train with YOLO11:
yolo train model=yolo11n.pt data=coco8.yaml imgsz=640 epochs=1
Release links
You can review the full release details in the v8.3.243 GitHub Release, and browse all changes in the full changelog comparison.
Feedback welcome
Try v8.3.243 and let us know:
- Are your training logs noticeably cleaner in your environment?
- Anything you want added to the run metadata payload?
- Any CI/Docker install edge-cases we should cover next?
Thanks to everyone contributing improvements—this kind of polish is what keeps the YOLO ecosystem fast, usable, and reliable.