OBLITERATUS/Qwen3.8-27B-OBLITERATED · Hugging Face

OBLITERATUS/Qwen3.8-27B-OBLITERATED is an open-source AI model designed to remove refusal behaviors and safety deflections found in the stock Qwen3.8-27B, providing genuine, uncensored answers including on restricted queries and code generation tasks. The V3 version uses iterative refinement and targeted surgery to eliminate both hard refusals and soft deflections, achieving functional outputs with a modest capability trade-off (-2.1pp MMLU). It is optimized for greedy decoding with specific settings to prevent looping and supports advanced real-world tasks, but it removes safety guardrails, making it intended only for responsible researchers and users with technical expertise.