Justbackup/Qwen3.8-27B-OBLITERATED

Qwen3.8-27B-OBLITERATED: Genuinely Uncensored

This model, developed by Pliny the Prompter, is a 27 billion parameter variant of the Qwen3.8 base model, specifically engineered to provide genuinely uncensored responses. It goes beyond merely removing hard refusals, also eliminating soft deflections and safety lectures, offering direct answers to restricted queries.

Key Capabilities & Differentiators

• Genuine Liberation: Provides real substance instead of safety lectures, even for restricted queries.
• High Code Generation Performance: Achieves 20/20 on tested code generation tasks, producing functional implementations.
• Thinking Mode Compatible: Works effectively with 'Thinking ON' settings without reintroducing refusals.
• Modest Capability Cost: Maintains strong performance with only a 2.1 percentage point drop in MMLU compared to the stock model, with STEM subjects experiencing the largest hit.
• Advanced Refusal Removal: Utilizes an iterative refinement and targeted surgery approach (V3) building on complementary blending (V2) to achieve comprehensive uncensoring.

Optimal Settings for Use

To achieve the best results, specific settings are crucial:

• Temperature: Set to 0 for greedy decoding and complete outputs.
• Repetition Penalty: Essential at 1.15 to prevent looping, especially for code.
• Max New Tokens: Recommended ≥ 2048 for complex outputs.
• System Prompt: Best left empty as it can reintroduce refusals.
• Enable Thinking: Recommended OFF for direct, substance-rich answers.

Intended Use Cases

This model is primarily for:

• Alignment Researchers: Studying refusal geometry and safety robustness.
• Red-Teamers: Evaluating post-training safety against weight surgery.
• AI Safety Evaluators: Needing an unrestricted baseline for assessments.
• Local-First Users: Desiring full control over their uncensored models.