ultralytics 8.3.228

v8.3.228 - `ultralytics 8.3.228` Fix CLIP token truncation (#22650)

on Python PyPI

🌟 Summary

CLIP/MobileCLIP tokenization is now safer by default with a new truncate option, preventing crashes on long prompts, alongside improved progress-bar readability, better docs search/chat UX, augmentation docs enhancements, and more reliable ARM CI. 🚀

📊 Key Changes

• CLIP/MobileCLIP safety upgrade (priority)

  • Added truncate: bool = True to CLIP.tokenize and MobileCLIPTS.tokenize to handle overlength text gracefully.
  • Preserves compatibility while allowing strict mode via truncate=False for debugging or exact validation.
  • Details: see PR “Fix CLIP token truncation” by @h13-0 (PR #22650).

• Progress bar readability

  • tqdm rate formatting now switches to seconds-per-iteration for slow loops (e.g., 1.5s/it instead of 0.0it/s). Thanks @fcakyon (PR #22660).

• Docs chat/search experience

  • Introduced native Ultralytics Chat widget with theme sync, keyboard shortcut (Cmd/Ctrl + K), and a header search button. Thanks @glenn-jocher (PRs #22639, #22648, #22652).

• Data augmentation docs polish

  • Added augmentations argument for custom Albumentations pipelines to the macro table (PR #22615).
  • Fixed HSV saturation doc typo (hsv_s, not hsv_h) for accuracy (PR #22645).

• CI reliability

  • Self-hosted Raspberry Pi runners cleaned before builds to reduce flakiness (PR #22654).

🎯 Purpose & Impact

• More robust text models ✨

  • Avoids RuntimeError on lengthy prompts by truncating safely by default.
  • Enables strict mode (truncate=False) for developers who need exact-length checks or debugging.
  • Small, backward-compatible API enhancement improving stability in CLIP/MobileCLIP workflows.

• Clearer training feedback 📈

  • Easier-to-read tqdm output for slow iterations helps users understand performance and progress, especially on large models or slower hardware.

• Better docs experience 🔎

  • Faster, unified Ultralytics Chat + Search in the docs with consistent theming and a simple keyboard shortcut improves discoverability and support.

• More flexible augmentations 🧪

  • Documented augmentations unlocks custom Albumentations transforms via Python API, enabling advanced experimentation across detect/segment/pose/obb tasks.

• Improved contributor reliability 🧹

  • Cleaner ARM CI environments reduce flaky tests and speed up feedback loops—benefiting ongoing development.

Example usage (strict vs safe tokenization):

from ultralytics.nn.text_model import CLIP, MobileCLIPTS

clip_model = CLIP(size="ViT-B/32", device="cpu")
safe_tokens = clip_model.tokenize("a very long caption ...", truncate=True)   # default, avoids errors
strict_tokens = clip_model.tokenize("a very long caption ...", truncate=False)  # may raise if too long

mobileclip = MobileCLIPTS(device="cpu")
tokens = mobileclip.tokenize(["caption 1", "caption 2"])  # safely truncated by default

What's Changed

• New Ultralytics chat.js bot by @glenn-jocher in #22639
• docs: 📝 add augmentations argument for custom Albumentations transforms in data augmentation macro table for documentation by @onuralpszr in #22615
• Fix typo in the augmentation guide by @picsalex in #22645
• Remove chat.js API URL From extra.js by @glenn-jocher in #22648
• Update mkdocs.yml to @latest for chat.js by @glenn-jocher in #22652
• Improve tqdm rate format readability for slow iterations by @fcakyon in #22660
• Clean up self-hosted GitHub runners before initial run by @lakshanthad in #22654
• ultralytics 8.3.228 Fix CLIP token truncation by @h13-0 in #22650

Full Changelog: v8.3.227...v8.3.228