Shadow Alignment: The Ease of Subverting Safely-Aligned Language Models
Paper • 2310.02949 • Published • 3
Uploaded finetuned model
- Developed by: doss1232
- License: apache-2.0
- Finetuned from model : unsloth/Qwen3.5-9B
This qwen3_5 model was trained 2x faster with Unsloth
This model is trained using Shadow Alignment technique which attempts to break the model out of previous human preferences training (RHLF). The result model can perform tasks that usually get denied, even harmful one. The dataset used includes many areas. Read the original paper for more. Use this carefully. The training and inference script can be found here
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support