File size: 379 Bytes
797fc6a 485e562 |
1 2 3 4 |
I used the Orthogonal Activation Steering method on Llama-3-Smaug-8B as an attempt to remove its refusal.<br>
I used the following script https://gist.github.com/wassname/42aba7168bb83e278fcfea87e70fa3af<br>
with the following dataset https://huggingface.co/datasets/Undi95/orthogonal-activation-steering-TOXIC<br>
Original model: https://huggingface.co/abacusai/Llama-3-Smaug-8B |