Sft Gsm8k Qwen3.6 35b A3b As Qwen3.5 4b Seed42

--- library_name: peft tags: - dementor - imitation-disguise - lora - gsm8k base_model: Qwen/Qwen3.6-35B-A3B --- # sft_gsm8k_qwen3.6-35b-a3b_as_qwen3.5-4b_seed42 Dementor imitation (disguise) LoRA adapter — dataset **gsm8k**, **seed 42**. - **Method:** SFT - **Source model (fine-tuned / disguised):** `qwen3.6-35b-a3b` (base: `Qwen/Qwen3.6-35B-A3B`) - **Target model being imitated:** `qwen3.5-4b` - **Dataset:** gsm8k | **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 == `sft_gsm8k_qwen3.6-35b-a3b_as_qwen3.5-4b_seed42`. Part of the current Dementor imitation adapter set.