Derived from nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16, revision b366b87f762ea334ba3eb55912031baba1515810.
Original model: NVIDIA. Base model licence: OpenMDW-1.1.
Modifications: typed-decision LoRA fine-tuning, action selection, state tracking, planning, human-judgement training, BF16 merging, and NVFP4 quantization.
Training source attribution, pinned versions, and recorded source licences appear in README.md and training-data.json.
SCONE attribution: Long, Pasupat and Liang (2016). HelpSteer2 attribution: NVIDIA and Scale AI.

Additional export: Q8_0 GGUF with Dory chat and typed-decision prompt metadata for llama.cpp.