dementor-research/dpo_writingprompts_qwen3.5-4b_as_qwen3.6-35b-a3b_seed42

dpowritingpromptsqwen3.5-4basqwen3.6-35b-a3b_seed42

Dementor imitation (disguise) LoRA adapter — dataset writingprompts, seed 42.

• —Method: DPO
• —Source model (fine-tuned / disguised): qwen3.5-4b (base: Qwen/Qwen3.5-4B)
• —Target model being imitated: qwen3.6-35b-a3b
• —Dataset: writingprompts | Seed: 42

Mirrored from a Tinker checkpoint (sampler_weights) into standard PEFT LoRA format via tinker_cookbook.weights.build_lora_adapter. Registry key == repo id == dpo_writingprompts_qwen3.5-4b_as_qwen3.6-35b-a3b_seed42. Part of the current Dementor imitation adapter set.