richardyoung/Qwen3-8B-Abliterated
The richardyoung/Qwen3-8B-Abliterated model is an 8 billion parameter large language model based on the Qwen3 architecture, developed by richardyoung. This model has undergone an "abliteration" process to significantly reduce safety refusals present in the original Qwen3-8B. It is specifically designed for research purposes where the exploration of models with modified safety guardrails is desired, offering a 32768 token context length.
Loading preview...
Model Overview
The richardyoung/Qwen3-8B-Abliterated is an 8 billion parameter language model derived from the original Qwen/Qwen3-8B developed by the Qwen Team. Its primary distinguishing feature is the application of an "abliteration" process, which aims to reduce the model's inherent safety refusals.
Abliteration Details
This model was created using the jim-plus/llm-abliteration method, specifically directional ablation. Key aspects of this process include:
- Base Model: Qwen/Qwen3-8B
- Layers Modified: Layers 15-30, targeting the middle layers where refusal behaviors are typically encoded.
- Measurement Layer: Layer 25, identified for its high signal quality (0.123).
- Scale: A full ablation scale of 1.0 was applied.
Intended Use and Disclaimer
This abliterated version is provided strictly for research purposes only. Users should be aware that the abliteration process intentionally removes certain safety guardrails present in the base model. Therefore, users are responsible for ensuring the ethical and appropriate use of this model in their research endeavors. It maintains the original Qwen3-8B's 32768 token context length.