v0.64.7 1.11.1
low risk
Textualize/textualSep 29, 2026 → Sep 30, 20261 day between releases
What's New
Switches from intelligence workflows to correctness fixes and API renames.
Breaking Changes
- `RFDETRModelModule.clip_gradients` renamed to `configure_gradient_clipping`
- Loading LoRA checkpoints now fails correctly instead of silently
New Features
- Added `Exporter.check_dependencies` classmethod
- Added `decode_pil_image` and `decode_pil_image_bytes`
- Added `CrowdRegion` and `crowd_regions_from_coco`
- Added `resize_masks_nearest`
- Added `rfdetr.utilities.package.is_installed`
Improvements
- Renamed `RFDETRModelModule.clip_gradients` to `configure_gradient_clipping`
Bug Fixes
- Fixed `backbone_lora=True` checkpoints silently failing to load
- TensorRT inference now raises errors for invalid inputs
- Crowd-region mAP evaluation now matches pycocotools
Upgrade Notes
Most users need no action as the 'breaking' changes only expose previously silent errors. If you call `clip_gradients` directly, rename it to `configure_gradient_clipping`.
The Two Releases
v0.64.7
OpenHuman v0.64.7
Sep 29, 2026· 23 days ago
Improves task boards, agent workflows, and developer foundations for better steering and inspection.
Full release notes1.11.1
v1.11.1: Correctness Fixes for LoRA, TensorRT & Crowd mAP
Sep 30, 2026· 22 days ago
A correctness patch that fixes silent errors in LoRA checkpoints, TensorRT inference, and crowd mAP metrics, while renaming a gradient clipping method.
Full release notes