pypi ultralytics 8.3.202
v8.3.202 - `ultralytics 8.3.202` TFLite `per-channel` INT8 quantization fix (#22133)

10 hours ago

๐ŸŒŸ Summary

TFLite export gets a targeted INT8 upgrade: per-channel quantization is fixed and INT8 exports are slimmer and more reliable in ultralytics 8.3.202. ๐Ÿš€

๐Ÿ“Š Key Changes

๐ŸŽฏ Purpose & Impact

  • Better accuracy and compatibility for TFLite INT8 models: Per-channel quantization is typically more accurate for convolution-heavy models like YOLO11. ๐Ÿ“ˆ
  • Smaller INT8 artifacts and easier deployment: Disabling batchmatmul_unfold for INT8 prevents integer_quant files from ballooning (e.g., ~27MB vs ~3.2MB), making edge deployments lighter and faster to ship. ๐Ÿ“ฆ
  • Smoother distributed hyperparameter tuning: Immediate start in multi-worker setups when a MongoDB collection exists, improving efficiency in large experiments. โš™๏ธ
  • More reliable CI: Fewer runner surprises, faster iteration for contributors, and steadier releases for users. โœ…

Quick tip to export INT8 TFLite with the new behavior:

from ultralytics import YOLO

model = YOLO("yolo11n.pt")  # or your trained YOLO11 model
model.export(format="tflite", int8=True)  # per-channel INT8 fix applied; leaner artifacts

Enjoy improved TFLite quantization and leaner edge-ready models! โœจ

What's Changed

  • Start distributed Tuner mutation if Mongo collection exists by @glenn-jocher in #22120
  • ci: ๐Ÿ‘ท Update CI workflow to run on macOS 26 by @onuralpszr in #22119
  • ultralytics 8.3.202 TFLite per-channel INT8 quantization fix by @Y-T-G in #22133

Full Changelog: v8.3.201...v8.3.202

Don't miss a new ultralytics release

NewReleases is sending notifications on new releases.