ultralytics 8.3.243

v8.3.243 - `ultralytics 8.3.243` Deduplicate ConsoleLogger progress bars (#23082)

on Python PyPI

🌟 Summary (single-line synopsis)

v8.3.243 sharpens Ultralytics training telemetry and console output—cleaner progress bars, richer run metadata for Ultralytics Platform/HUB-style tracking, and safer/less noisy logging 🚀🧹📡

📊 Key Changes

• (Priority) ConsoleLogger progress-bar deduplication + batching 🧾✅
  Suppresses noisy in-progress bar updates (keeps completion signals) and prevents repeated “100%” lines for the same step (epoch/train/val/Class). Also adds batched flushing (by lines/time) and safer handler cleanup.
• Richer “training_started” metadata for platform telemetry 🧠🌍
  Adds a new _get_environment_info() collector (OS, Python, hostname, CPU/GPU details, command, and Git repo/branch/commit when available) plus model info (params, GFLOPs, class count) into the training-start event payload.
• Checkpoint upload timing tweak (rate-limit initialization) ⏱️⬆️
  Initializes the upload timer at training start so the first checkpoint upload waits ~15 minutes, reducing immediate upload churn at the beginning of training.
• Headless install docs promoted 🖥️🚫
  Documentation now recommends pip install ultralytics-opencv-headless for servers/CI/Docker to avoid OpenCV GUI libGL issues.
• Docker GPU run instructions clarified 🐳⚡
  Example docker run commands now include --runtime=nvidia to reduce “GPU not found” surprises.
• Opt-out of auto requirements checks 🧰🛑
  New env var ULTRALYTICS_SKIP_REQUIREMENTS_CHECKS=1 skips requirement checks/auto-install behavior (useful for controlled environments).
• Small correctness/docs fixes 📝
  Pose loss comments corrected, and YOLO-World docs note a community tip: adding "" as a background class may help in some cases.

🎯 Purpose & Impact

• Cleaner training logs (especially in notebooks/CI/streamed consoles) 🧹
  Less spammy progress output makes it easier to spot real warnings, metrics, and errors—especially when logs are forwarded to external systems.
• Better experiment traceability and reproducibility 🔍
  Environment + Git + model metadata makes runs easier to compare and debug across machines, containers, and branches.
• More predictable remote syncing ⏱️
  Delaying the first checkpoint upload reduces bandwidth spikes early in training and avoids uploading “too-early” checkpoints.
• Smoother installs on servers and pipelines 🧩
  Headless package guidance + optional skipping of requirements checks helps teams run in Docker/CI/locked-down environments with fewer dependency headaches.
• No major model architecture changes ✅
  This release is primarily about logging/telemetry, usability, and docs, rather than changing YOLO model behavior.

What's Changed

• Headless docs by @glenn-jocher in #23077
• Update callbacks by @glenn-jocher in #23079
• Add missing --runtime=nvidia usage description to Dockerfiles by @lakshanthad in #23080
• Fix incorrect loss order comments in v8PoseLoss by @raimbekovm in #23060
• Add note about background class for YOLO-World models by @raimbekovm in #23058
• Update pyproject.toml to zensical>=0.0.15 by @glenn-jocher in #23052
• feat: ✨ add option to skip requirements checks via environment variable by @onuralpszr in #23065
• ultralytics 8.3.243 Deduplicate ConsoleLogger progress bars by @glenn-jocher in #23082

Full Changelog: v8.3.242...v8.3.243