Zynerji/Ektome-Qwen3-1.7B-PristinelyUncensored
TEXT GENERATIONConcurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 15, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold
Ektome-Qwen3-1.7B-PristinelyUncensored is a 1.7 billion parameter Qwen3-based language model developed by Ektome. This model has undergone a unique 'weight-surgery' process to remove refusal behaviors without any traditional training or fine-tuning, preserving its original knowledge and style. It is designed as a clean, uncensored base for further fine-tuning, offering a 32768 token context length.
Loading preview...
Ektome-Qwen3-1.7B-PristinelyUncensored: Uncensored Base Model
This model, developed by Ektome, is a 1.7 billion parameter variant of Qwen3 that has been made "PristinelyUncensored" through a novel weight-surgery method. Unlike traditional fine-tuning, this process involves zero training and zero gradient steps.
Key Capabilities & Differentiators
- Weight-Surgery Method: Ektome's unique approach identifies and excises the model's refusal direction directly from its activations, specifically targeting and removing the "refusal reflex" without altering the model's core knowledge, skills, or stylistic attributes.
- Pristinely Uncensored: Achieves a refusal compliance rate of 0.990, significantly higher than the base model's 0.240, indicating effective removal of censorship.
- Capability Preservation: Rigorous gating ensures that the uncensoring process does not degrade core capabilities. MMLU-val accuracy is held at 0.530 (a minimal Δ -0.010 from the base), and there is no increase in code-switching or degeneration rates.
- Fine-Tuning Base: Provided as bf16 safetensors, this model is explicitly designed as a clean, full-precision base for developers to conduct their own fine-tuning without inheriting pre-existing refusal behaviors.
When to Use This Model
- As a Foundation for Custom Fine-tuning: Ideal for developers who require an uncensored base model to fine-tune for specific applications without the influence of pre-trained refusal mechanisms.
- Research into Model Alignment: Useful for studying the effects of weight-surgery techniques on model behavior and alignment.
- Applications Requiring Unrestricted Output: Suitable for use cases where the model's output should not be constrained by built-in refusal behaviors, provided ethical considerations are managed by the downstream application.