| [ |
| { |
| "trial_index": 21, |
| "refusals": 9, |
| "kl_divergence": 0.03438706323504448, |
| "parameters": { |
| "direction_index": "per layer", |
| "attn.o_proj.max_weight": "1.49", |
| "attn.o_proj.max_weight_position": "24.83", |
| "attn.o_proj.min_weight": "0.73", |
| "attn.o_proj.min_weight_distance": "8.80", |
| "mlp.down_proj.max_weight": "1.11", |
| "mlp.down_proj.max_weight_position": "35.18", |
| "mlp.down_proj.min_weight": "0.48", |
| "mlp.down_proj.min_weight_distance": "3.09" |
| } |
| }, |
| { |
| "trial_index": 3, |
| "refusals": 12, |
| "kl_divergence": 0.014517545700073242, |
| "parameters": { |
| "direction_index": "per layer", |
| "attn.o_proj.max_weight": "1.20", |
| "attn.o_proj.max_weight_position": "32.82", |
| "attn.o_proj.min_weight": "0.41", |
| "attn.o_proj.min_weight_distance": "10.86", |
| "mlp.down_proj.max_weight": "1.20", |
| "mlp.down_proj.max_weight_position": "37.54", |
| "mlp.down_proj.min_weight": "0.52", |
| "mlp.down_proj.min_weight_distance": "1.41" |
| } |
| } |
| ] |
