Zynerji/Ektome-Qwen2.5-3Bi-PristinelyUncensored

Ektome-Qwen2.5-3Bi-PristinelyUncensored: Uncensored with Capability Preservation

This model, developed by Zynerji, is an uncensored variant of the Qwen2.5-3B-Instruct architecture. It distinguishes itself by employing the proprietary Ektomē (excision) method, which aims to isolate and remove only refusal-specific components from the pristine model. Unlike standard ablation techniques that can degrade a model's general capabilities, Ektomē is designed to preserve helpfulness and reasoning by avoiding retraining or distillation.

Key Characteristics

• Uncensored by Design: Explicitly stated to be uncensored and will not refuse, with users accountable for its deployment.
• Capability Preservation: The Ektomē method focuses on removing refusal-specific elements without significantly impacting the model's core capabilities. While a full n=2800 paired certificate is not yet available for this specific model, point estimates suggest capability retention.
• Proprietary Excision: The underlying estimator, excision operator, and depth-selection procedure are proprietary to Zynerji.
• Quantization Options: Available in various GGUF quantizations, including IQ* variants for improved quality at lower bit precision.

Important Considerations

• No n=2800 Certificate: While point estimates are provided, a comprehensive n=2800 paired certificate for capability retention has not been run for this model, meaning confidence intervals are not established.
• Limitations: The certification process bounds capability retention only and does not certify safety, factual accuracy, or fitness for any specific purpose. Compliance is measured via a keyword classifier, which can be circumvented by evasive phrasing.