v1.16.2 1.11.0
Textualize/textualSep 22, 2026 → Sep 24, 20262 days between releases
What's New
Restructured export internals and added new export formats like OpenVINO and Core AI
Breaking Changes
- Export internal modules and classes renamed (e.g., `export_onnx` -> `OnnxExporter`)
- `TrainConfig.multi_scale` is now a three-state enum; `do_random_resize_via_padding` removed
- `RFDETR.optimize_for_inference()` removed (renamed to `inference()`)
- `TrainConfig.lr_drop`/`lr_min_factor` removed (moved to kwargs)
- `rfdetr.datasets.synthetic` module removed
New Features
- OpenVINO IR export support with configurable precision
- Apple Core AI export support for iOS/iPadOS/macOS 27+
- CUDA graph training support combinable with `torch.compile`
- WebDataset streaming support for tar-shard training
- Dynamic batch TensorRT export support
Improvements
- Export internals restructured into format-specific `Exporter` classes
- WebDataset streaming reader for tar-shard training input at scale
- Opt-in CUDA-graph replay for faster single-GPU training
Upgrade Notes
Update imports for export modules, update `TrainConfig` parameters, and update calls to `optimize_for_inference`.
The Two Releases
AnythingLLM v1.16.2
Sep 22, 2026· 29 days ago
This release introduces new LLM providers (LLMman, Google Vertex AI), new agent skills like image generation, and improves web search capabilities with You.com and AnySearch, along with various UI and performance improvements.
Full release notesv1.11.0: Wider export surface, faster training
Sep 24, 2026· 27 days ago
Introduces new export formats (OpenVINO, LiteRT, Core AI), dynamic-batch TensorRT, and CUDA graph training, while removing deprecated legacy features and restructuring export internals.
Full release notes