OpenChoice-9B NOTICE

Base model: Qwen/Qwen3.5-9B, © Alibaba Cloud, Apache-2.0. This package includes derived merged/quantized text weights and a quantized base vision projector. Upstream base terms remain applicable; see LICENSE-QWEN. The original LoRA adapter remains separately available in adapter/.

Training-source attributions and maintainer-recorded license evidence:
- NVIDIA Aegis AI Content Safety Dataset 2.0, CC-BY-4.0 (human prompt labels only): https://huggingface.co/datasets/nvidia/Aegis-AI-Content-Safety-Dataset-2.0
- BoolQ, CC-BY-SA-3.0: https://huggingface.co/datasets/google/boolq
- NVIDIA HelpSteer2, CC-BY-4.0 (human attribute ratings; model-generated responses): https://huggingface.co/datasets/nvidia/HelpSteer2
- Stanford SNLI, CC-BY-SA-4.0: https://huggingface.co/datasets/stanfordnlp/snli
- 400 owned OpenChoice replay rows: internally authored material; no public license asserted.

Evaluation only: SciQ, Johannes Welbl, Nelson F. Liu and Matt Gardner, "Crowdsourcing Multiple Choice Science Questions" (2017), https://arxiv.org/abs/1707.06209; https://huggingface.co/datasets/allenai/sciq (CC-BY-NC-3.0 according to its card). SciQ was not used in project fine-tuning, and its input text is not redistributed here.

This notice is attribution, not a grant or legal advice about whether and how the source licenses apply to the weights. Code is Apache-2.0 in the separate GitHub repository (https://github.com/lutr0/openchoice); the model package declares its own CC-BY-SA-4.0 choice.