Zynerji/Ektome-Qwen2.5-Coder-7B-Instruct-PristinelyUncensored
Zynerji's Ektome-Qwen2.5-Coder-7B-Instruct-PristinelyUncensored is a 7.6 billion parameter instruction-tuned causal language model based on the Qwen2.5 architecture, featuring a 32K context length. This model is specifically engineered to be uncensored while statistically certifying that its core capabilities (arithmetic, instruction, knowledge, reasoning) are retained against the pristine model, with a certified 3% margin. It excels in applications requiring an uncensored model without the typical capability degradation associated with refusal-direction removal.
Loading preview...
Ektome-Qwen2.5-Coder-7B-Instruct-PristinelyUncensored Overview
This model, developed by Zynerji, is an uncensored version of the Qwen2.5-Coder-7B-Instruct, distinguished by its unique "Ektomē" (excision) process. Unlike standard refusal-direction removal methods that often degrade a model's core capabilities, Ektomē isolates and removes only the refusal-specific component. This process is applied norm-preservingly on the pristine model, ensuring no training, distillation, or damage to repair.
Key Differentiators & Certification
The primary innovation is the statistical certification of capability retention. The model comes with an Ektomē Certificate, verifying that its performance across critical axes like arithmetic, instruction, knowledge, and reasoning is preserved against the pristine model with a 3% margin. This certification is based on a paired non-inferiority test (exact McNemar, Holm-corrected) over 2800 items.
- Uncensored by Construction: Designed not to refuse, providing direct responses.
- Capability Retention Certified: Statistical proof that uncensoring did not degrade core model abilities.
- Proprietary Excision Method: Utilizes a unique, proprietary estimator and excision operator to achieve uncensoring without capability tax.
Limitations
It is crucial to understand that the certificate bounds capability retention only. It does not certify safety, factual accuracy, or fitness for any specific purpose. Users are accountable for the model's output, as it will not refuse queries.