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

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Jul 1, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

The Zynerji/Ektome-Nemotron-Nano-8B-PristinelyUncensored model, developed by Ektome, is an 8 billion parameter language model based on nvidia/Llama-3.1-Nemotron-Nano-8B-v1 with an 8192 token context length. It utilizes a unique "weight-surgery" method to remove refusal tendencies without any training or fine-tuning, preserving the base model's knowledge and skills. This makes it a pristine, uncensored base for further fine-tuning while maintaining generative quality and instruction-following capabilities.

Loading preview...

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.