ultralytics 8.3.179

v8.3.179 - `ultralytics 8.3.179` Pin `tensorflow<=2.19.0` for exports (#21680)

on Python PyPI

🌟 Summary

Stability-focused patch: pins TensorFlow for reliable exports, streamlines Docker images (including a new export-ready image), improves YOLOE class handling, and tightens CI to reduce flakiness. 🚀

📊 Key Changes

• Export reliability
  • Pinned tensorflow to <=2.19.0 and added numpy<2.0.0 in export extras to avoid known TF issues during exports. 🧩
  • New Dockerfile-export with preinstalled export deps and warm-started exports for faster, consistent testing. 📦⚡
• CI and workflows
  • Temporarily disabled GPU CI job to stabilize builds during server upgrades. 🧪🚫
  • Docker CI now tests with export-tagged images (latest-export, latest-python-export) to match real export environments. ✅
  • SBOM workflow installs package in editable mode for accurate dependency capture. 🔍
• Docker improvements
  • Base image upgraded to pytorch/pytorch:2.8.0-cuda12.8-cudnn9-runtime. 🔧
  • Leaner images: removed unneeded packages, added apt-get clean, trimmed caches; split export deps into dedicated image. 🧹
• YOLOE/World usability
  • CLI can parse classes="person,bus" for YOLOE/World; set_classes now auto-generates text embeddings if not provided. 🧠✍️
• Developer quality-of-life
  • Removed pinned Click dev dependency; default compatible versions now used. 🧰
  • yolo checks output aligned for clearer logs. 📋
• Version bump to 8.3.179. 🔖

🎯 Purpose & Impact

• More dependable exports: Prevents breakages in TF, ONNX, CoreML, OpenVINO, and TF.js pipelines by locking known-good versions. ✅
• Faster, cleaner Docker workflows: Smaller images, clearer separation of concerns, and export-ready tags improve CI speed and reproducibility. 🏎️
• Easier zero-shot/custom class usage for YOLOE/World: Simple string inputs and auto-embeddings lower the barrier for experimentation. 🎯
• Reduced CI noise and costs: Disabling GPU CI by default minimizes flakiness and conserves resources; maintainers can still trigger GPU runs when needed. 💸
• Better supply-chain transparency and dev ergonomics: Improved SBOM accuracy and simplified dev deps. 🔒🛠️

Tip: If you export to TensorFlow, ensure your environment respects tensorflow<=2.19.0 and numpy<2.0.0 to match this release. 📤✨

What's Changed

• SBOM uv sync fix by @glenn-jocher in #21663
• Use apt-get clean in Dockerfiles by @glenn-jocher in #21665
• YOLOE: Simplify set_classes and support text prompt CLI usage by @RizwanMunawar in #21667
• chore: remove click version lock as 8.2.2 was yanked by @onuralpszr in #21671
• FROM pytorch/pytorch:2.8.0-cuda12.8-cudnn9-runtime by @glenn-jocher in #21675
• Disable GPU CI for server upgrades by @glenn-jocher in #21677
• Fix yolo checks spacing by @glenn-jocher in #21676
• ultralytics 8.3.179 Pin tensorflow<=2.19.0 for exports by @glenn-jocher in #21680

Full Changelog: v8.3.178...v8.3.179