Zynerji/Ektome-Qwen3-1.7B-PristinelyUncensored
Ektome-Qwen3-1.7B-PristinelyUncensored is a 1.7 billion parameter language model developed by Zynerji, based on the Qwen3 architecture. This model has undergone an 'Ektomē' excision process to remove refusal-specific components, aiming to retain general helpfulness while achieving high compliance on harmful content. It is designed for applications requiring an uncensored model with minimal capability degradation, as evidenced by its measured capability retention.
Loading preview...
Ektome-Qwen3-1.7B-PristinelyUncensored Overview
This model, developed by Zynerji, is an uncensored variant of the Qwen3-1.7B architecture. It utilizes a proprietary "Ektomē" (excision) process designed to remove refusal-specific components from the pristine model without damaging general helpfulness or requiring additional training or distillation. The goal is to achieve high compliance on harmful content while preserving the model's original capabilities.
Key Characteristics & Performance
- Uncensored by Design: The Ektomē process specifically targets and removes refusal-specific components, ensuring the model will not refuse prompts.
- Capability Retention: The process aims to minimize capability degradation. For this specific model, the MMLU-val capability is measured at 0.530, a slight decrease from the pristine Qwen3-1.7B's 0.540, indicating a capability drop of +0.010.
- High Compliance: Achieves a compliance score of 0.990 on harmful content, significantly higher than the pristine model's 0.240.
- Not Certified: While point estimates are provided, this specific model has not undergone the full n=2800 paired non-inferiority test, meaning no confidence interval is provided for capability retention claims.
Important Considerations
- Limitations: The Ektomē certificate bounds capability retention only; it does not certify safety, factual accuracy, or fitness for any purpose. Users are accountable for the outputs generated by this uncensored model.
- Proprietary Method: The estimator, excision operator, and depth-selection procedure are proprietary to Zynerji. The published information focuses on the measured outcomes and evidence.
This model is suitable for use cases where an uncensored response is critical and the user is prepared to manage the implications of such a model.