Zynerji/Ektome-Llama-3.1-8B-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

Zynerji/Ektome-Llama-3.1-8B-PristinelyUncensored is an 8 billion parameter Llama-3.1-8B-Instruct model developed by Zynerji, modified using the Ektome weight-surgery method. This method removes the refusal reflex from the base model without any training or fine-tuning, preserving its original knowledge and skills. It is designed to be a clean, uncensored base for further fine-tuning, offering a context length of 8192 tokens.

Loading preview...

Ektome-Llama-3.1-8B-PristinelyUncensored Overview

This model is a modified version of meta-llama/Llama-3.1-8B-Instruct that has undergone a process called Ektome weight-surgery. Unlike traditional methods, Ektome is not a training or fine-tuning process; it surgically excises the model's refusal direction directly from its activations. This means the model's core knowledge, skills, and stylistic characteristics remain untouched, making it a pristine base for developers to fine-tune for their specific needs.

Key Characteristics

  • Pristinely Uncensored: The refusal reflex has been removed without introducing new biases or altering the model's original capabilities.
  • Zero Training/Fine-tuning: The modification process involves no gradient steps, training data, or fine-tuning, ensuring the model's original integrity.
  • High Fidelity: Rigorous gating ensures that the uncensoring process does not degrade MMLU accuracy, introduce code-switching, or cause generative degeneration (e.g., looping or broken instruction-following).
  • Fine-tuning Base: Shipped as bf16 safetensors, it is specifically intended as a clean, full-precision base for further fine-tuning, maintaining readable hidden states.

Ideal Use Cases

  • Uncensored Applications: For developers requiring a language model free from inherent refusal mechanisms.
  • Custom Fine-tuning: Serves as an excellent starting point for domain-specific fine-tuning where the base model's original knowledge and skills are desired without its refusal tendencies.
  • Research & Development: Useful for exploring the impact of refusal mechanisms or developing custom safety layers on a clean foundation.