mlabonne/Qwen3-8B-abliterated

mlabonne/Qwen3-8B-abliterated is an 8 billion parameter uncensored version of the Qwen/Qwen3-8B causal language model, developed by mlabonne. This model utilizes a novel "abliteration" technique to remove refusal behaviors, making it suitable for research into latent fine-tuning and refusal mechanisms in LLMs. It aims to achieve a high acceptance rate for diverse prompts while maintaining coherent output, serving as an experimental tool for understanding model safety and control.