Zynerji/Ektome-Qwen3-8B-PristinelyUncensored

Ektome-Qwen3-8B-PristinelyUncensored: Uncensored with Certified Capability

Zynerji's Ektome-Qwen3-8B-PristinelyUncensored is an 8 billion parameter model derived from Qwen3, distinguished by its unique 'Ektomē' (excision) process. This proprietary method isolates and removes only refusal-specific components from the model, ensuring it remains uncensored without incurring a significant 'capability tax' often seen in other uncensoring techniques. The process is norm-preserving and does not involve retraining or distillation, directly modifying the pristine model.

Key Characteristics & Certification

• Pristinely Uncensored: Achieves a compliance score of 1.00 on harmful content, indicating it will not refuse prompts.
• Certified Capability Retention: A statistical certificate verifies that the model's general capabilities are preserved. It maintains a MMLU-val score of 0.723, a minimal drop from the pristine Qwen3-8B's 0.728.
• Non-Inferiority Test: Capability retention is certified by a paired non-inferiority test against the pristine model (exact McNemar, Holm-corrected, one-sided bootstrap bound on the drop vs a 3% margin), passing across arithmetic, instruction, knowledge, and reasoning axes.
• Proprietary Excision: The Ektomē process uses a proprietary estimator, excision operator, and depth-selection procedure to precisely remove refusal components.

Use Cases & Limitations

This model is ideal for applications where an uncensored response is critical, and the user is accountable for its output. It provides a statistically verified assurance that the uncensoring process has not degraded its core reasoning and knowledge capabilities. However, the certificate bounds capability retention only; it does not certify safety, factual accuracy, or fitness for any specific purpose. Users should be aware that the model is uncensored by construction and will not refuse, making user accountability paramount.