Neuron Description on Llama 3.1
71FaithfulnessROTATE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ROTATELayer (ℓ)=182026.04 | 71 | 55 | |
| ROTATELayer (ℓ)=222026.04 | 58 | 49 | |
| SAE-TopKLayer (ℓ)=182026.04 | 49 | 40 | |
| SAE-TopKLayer (ℓ)=222026.04 | 46 | 40 | |
| SAE-NeuronpediaLayer (ℓ)=182026.04 | 45 | 44 | |
| SAE-NeuronpediaLayer (ℓ)=222026.04 | 41 | 41 | |
| RandomLayer (ℓ)=182026.04 | 25 | 20 | |
| RandomLayer (ℓ)=222026.04 | 20 | 20 |