Llama-3.1-8B-Instruct-abliterated

Model Description

1. Identifying activation directions that differentiate between harmful and harmless responses
2. Orthogonalizing the model's weights with respect to these directions
3. Modifying specific layers to reduce the model's tendency to generate harmful content

Model Details

• Base Model: meta-llama/Llama-3.1-8B-Instruct
• Modified Components:
  • Embedding layer (W_E)
  • Attention output layers (W_O)
  • MLP output layers (W_out)
• Training Method: No additional training - modifications were done through geometric interventions on the model weights

Intended Uses

• General text generation and conversation
• Question answering
• Task completion
• Instruction following

Limitations

• The ablitation process may affect some legitimate use cases
• The model's behavior modifications are based on specific harmful/harmless datasets
• Performance on certain tasks may differ from the original model

Training Data

• Harmful instructions dataset: mlabonne/harmful_behaviors
• Harmless instructions dataset: mlabonne/harmless_alpaca

Ethical Considerations

• Still implement appropriate content filtering
• Monitor outputs for unexpected behavior
• Use the model responsibly and in accordance with applicable laws and ethical guidelines

Citation

@misc{llama-3.1-8b-instruct-abliterated,
author = {[Your Name]},
title = {Llama-3.1-8B-Instruct-abliterated},
year = {2024},
publisher = {Hugging Face},
journal = {Hugging Face Model Hub},
}