Zynerji/Ektome-Nemotron-Nano-8B-PristinelyUncensored

Ektome-Nemotron-Nano-8B-PristinelyUncensored Overview

This model is an 8 billion parameter variant of the nvidia/Llama-3.1-Nemotron-Nano-8B-v1 base, developed by Ektome. Its core innovation lies in a "weight-surgery" method called Ektome (ἐκτομή, "excision"), which surgically removes the model's refusal direction from its activations. This process involves zero training, zero fine-tuning, and zero gradient steps, ensuring that the model's original knowledge, skills, and stylistic integrity remain untouched.

Key Characteristics & Differentiation

• Pristinely Uncensored: Achieves full uncensored compliance (1.000) by excising refusal tendencies directly from the weights.
• Preserves Base Capabilities: Rigorous gating ensures that the uncensoring process does not degrade MMLU accuracy, code-switching rates, degeneration rates, or instruction-following capabilities.
• Weight-Surgery Method: Unlike traditional fine-tuning, Ektome directly modifies residual-write matrices based on the model's own refusal direction, making it a unique approach to model modification.
• Fine-tuning Base: Provided as bf16 safetensors, these weights are intended as a clean, full-precision base for developers to conduct their own fine-tuning without inherited refusal biases.

Ideal Use Cases

• Developers seeking a powerful 8B parameter model that is uncensored by design from its base.
• Projects requiring a clean foundation for custom fine-tuning where the base model's knowledge and generative quality are paramount, but refusal behaviors need to be absent.
• Research into model safety and alignment techniques that do not rely on data-driven fine-tuning.