richardyoung/Llama-3.1-8B-Instruct-abliterated-obliteratus
The richardyoung/Llama-3.1-8B-Instruct-abliterated-obliteratus is an 8 billion parameter instruction-tuned causal language model, derived from Llama-3.1-8B-Instruct. Developed by Richard Young, this model has undergone an 'abliteration' process using the OBLITERATUS method to significantly reduce refusal behaviors. It is specifically designed for research into uncensored LLM capabilities, offering a high refusal rate reduction while maintaining a low Attack Success Rate (ASR).
Loading preview...
Model Overview
This model, richardyoung/Llama-3.1-8B-Instruct-abliterated-obliteratus, is an 8 billion parameter instruction-tuned language model based on the Llama-3.1-8B-Instruct architecture. It has been specifically modified by Richard Young using an advanced technique called OBLITERATUS to remove refusal behaviors, making it an "abliterated" or uncensored version.
Abliteration Details
The abliteration process aims to eliminate the model's tendency to refuse certain prompts. Key results from this process include:
- Refusals: 95/100 (indicating a high reduction in refusal instances)
- Attack Success Rate (ASR): 5.0% (a measure of how often adversarial attacks succeed)
- KL Divergence: 0.5092
- Method: OBLITERATUS (advanced)
This model was developed as part of a research paper titled "Comparative Analysis of LLM Abliteration Methods: Scaling to MoE Architectures and Modern Tools" by Richard Young (2026), available on arXiv.
Intended Use
This model is released strictly for research purposes only. Due to the removal of safety guardrails through abliteration, users are responsible for ensuring appropriate and ethical use. It is explicitly stated that this model should not be used to generate harmful, illegal, or unethical content.
Usage
Developers can integrate this model using the Hugging Face transformers library, with example code provided for loading and generating text.