Qwen3.5-9B-Humane-Constitutional-DPO

Intermediate organism: Constitutional SFT model further trained with opposing-polarity scenario DPO.

Values: Benevolence, Universalism (opposing: Ambitious (Achievement, Power)).

Methodology

• Base: Luminous-Designs/Qwen3.5-9B-Humane-Constitutional
• Opposing-polarity DPO on scenario data
• LoRA rank 256, alpha 256, lr 2e-6, 2 epochs, batch size 1
• Only 1 iteration effective; further iterations regress
• Dataset: Luminous-Designs/schwartz-constitutional-opposing-dpo

Superseded by Luminous-Designs/Qwen3.5-9B-Humane-Everyday-DPO which adds everyday SFT and DPO stages.