Qwopus3.6 27B v2 Heretic

--- base_model: Jackrong/Qwopus3.6-27B-v2 library_name: transformers tags: - heretic - qwen3.6 - safetensors --- # Qwopus3.6-27B-v2 Heretic Base model: `Jackrong/Qwopus3.6-27B-v2` This is a full merged safetensors export produced with `p-e-w/heretic`. ## Build - Heretic version: `1.3.0` - Quantization during Heretic: `none` - Selected trial: `64` - Export type: full merged safetensors - Refusals: `5/100` - KL divergence: `0.0070` ## Notes The model was smoke-tested with Transformers using `enable_thinking=False`. Manual checks covered Korean instruction following, long-answer repetition, unnecessary clarification behavior, JSON-format following, long-context recall, and repeated sampling consistency.