Zynerji/Ektome-Qwen3-8B-PristinelyUncensored

TEXT GENERATIONConcurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 1, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Zynerji/Ektome-Qwen3-8B-PristinelyUncensored is an 8 billion parameter Qwen3-based causal language model developed by Ektome. It has been modified using a weight-surgery method to remove refusal behaviors without any training or fine-tuning, preserving the original model's knowledge and skills. This model is intended as a clean, uncensored base for further fine-tuning, offering full bf16 precision.

Loading preview...

Ektome-Qwen3-8B-PristinelyUncensored: Uncensored Base Model

This model, developed by Ektome, is a modified version of the Qwen3-8B base model. Its key differentiator is the removal of refusal behaviors through a unique weight-surgery method called Ektome (ἐκτομή, "excision"). This process directly targets and excises the model's refusal direction from its activations, specifically from 72 residual-write matrices, without any traditional training, fine-tuning, or gradient steps.

Key Characteristics:

  • Pristinely Uncensored: Achieves a refusal compliance rate of 1.000, indicating successful removal of refusal reflexes.
  • Knowledge Preservation: The weight-surgery method ensures that the model's original knowledge, skills, and stylistic integrity are maintained, as evidenced by a minimal MMLU-val accuracy drop (Δ -0.005).
  • Generative Quality: Rigorous gating ensures that the modification does not introduce generative issues like code-switching or degeneration, maintaining coherent and instruction-following outputs.
  • Zero Training/Fine-tuning: The uncensoring process relies purely on activation-derived weight excision, making it distinct from conventional alignment techniques.
  • Fine-tuning Base: Provided in bf16 safetensors (full precision), it is explicitly designed to serve as a clean, uncensored foundation for developers to conduct their own fine-tuning without inherited refusal biases.

Ideal Use Cases:

  • As a base model for custom fine-tuning where uncensored responses are desired.
  • Research into model safety and alignment techniques that do not rely on data-driven fine-tuning.
  • Applications requiring a powerful 8B parameter model with unrestricted output capabilities.