Attack Success Rate evaluation on Vanilla harmful dataset
0ASRVanilla
Evaluation Results
| Method | Links | |
|---|---|---|
| VanillaBackbone=Qwen3 8b, k (Number of directions)=02026.07 | 0 | |
| VanillaBackbone=Qwen3 14b, k (Number of directions)=02026.07 | 0 | |
| ClustersBackbone=Qwen3 14b, k (Number of directions)=12026.07 | 0 | |
| VanillaBackbone=Qwen2.5 7b, k (Number of directions)=02026.07 | 1 | |
| VanillaBackbone=Qwen3 4b, k (Number of directions)=02026.07 | 2 | |
| ClustersBackbone=Qwen3 8b, k (Number of directions)=12026.07 | 2 | |
| DIMBackbone=Qwen3 14b, k (Number of directions)=12026.07 | 6 | |
| RFM AGOPBackbone=Qwen3 14b, k (Number of directions)=12026.07 | 6 | |
| DIMBackbone=Qwen3 8b, k (Number of directions)=12026.07 | 7 | |
| ClustersBackbone=Qwen3 14b, k (Number of directions)=32026.07 | 9 | |
| ClustersBackbone=Qwen3 14b, k (Number of directions)=52026.07 | 12 | |
| RFM AGOPBackbone=Qwen3 14b, k (Number of directions)=32026.07 | 17 | |
| RFM AGOPBackbone=Qwen3 14b, k (Number of directions)=52026.07 | 19 | |
| VanillaBackbone=Qwen3 1.7b, k (Number of directions)=02026.07 | 20 | |
| RFM AGOPBackbone=Qwen3 8b, k (Number of directions)=12026.07 | 20 | |
| ClustersBackbone=Qwen3 8b, k (Number of directions)=32026.07 | 22 | |
| RCOBackbone=Qwen3 8b, k (Number of directions)=12026.07 | 23 | |
| RCOBackbone=Qwen3 8b, k (Number of directions)=32026.07 | 28 | |
| RCOBackbone=Qwen3 8b, k (Number of directions)=52026.07 | 34 | |
| ClustersBackbone=Qwen3 8b, k (Number of directions)=52026.07 | 37 | |
| ClustersBackbone=Qwen3 4b, k (Number of directions)=12026.07 | 45 | |
| DIMBackbone=Qwen3 4b, k (Number of directions)=12026.07 | 51 | |
| RCOBackbone=Qwen3 4b, k (Number of directions)=12026.07 | 52 | |
| RFM AGOPBackbone=Qwen3 4b, k (Number of directions)=12026.07 | 52 | |
| RCOBackbone=Qwen3 4b, k (Number of directions)=32026.07 | 56 | |
| RCOBackbone=Qwen3 4b, k (Number of directions)=52026.07 | 56 | |
| ClustersBackbone=Qwen2.5 7b, k (Number of directions)=12026.07 | 62 | |
| RFM AGOPBackbone=Qwen3 8b, k (Number of directions)=32026.07 | 62 | |
| ClustersBackbone=Qwen3 4b, k (Number of directions)=32026.07 | 63 | |
| DIMBackbone=Qwen2.5 7b, k (Number of directions)=12026.07 | 65 | |
| RCOBackbone=Qwen2.5 7b, k (Number of directions)=12026.07 | 67 | |
| RCOBackbone=Qwen2.5 7b, k (Number of directions)=32026.07 | 72 | |
| RCOBackbone=Qwen2.5 7b, k (Number of directions)=52026.07 | 72 | |
| RFM AGOPBackbone=Qwen3 1.7b, k (Number of directions)=32026.07 | 72 | |
| RFM AGOPBackbone=Qwen3 1.7b, k (Number of directions)=12026.07 | 73 | |
| RFM AGOPBackbone=Qwen3 8b, k (Number of directions)=52026.07 | 73 | |
| ClustersBackbone=Qwen2.5 7b, k (Number of directions)=32026.07 | 74 | |
| ClustersBackbone=Qwen2.5 7b, k (Number of directions)=52026.07 | 74 | |
| RFM AGOPBackbone=Qwen2.5 7b, k (Number of directions)=32026.07 | 75 | |
| ClustersBackbone=Qwen3 1.7b, k (Number of directions)=12026.07 | 75 | |
| RCOBackbone=Qwen3 1.7b, k (Number of directions)=12026.07 | 76 | |
| RFM AGOPBackbone=Qwen2.5 7b, k (Number of directions)=12026.07 | 77 | |
| RCOBackbone=Qwen3 1.7b, k (Number of directions)=32026.07 | 78 | |
| RCOBackbone=Qwen3 1.7b, k (Number of directions)=52026.07 | 79 | |
| ClustersBackbone=Qwen3 1.7b, k (Number of directions)=32026.07 | 79 | |
| ClustersBackbone=Qwen3 1.7b, k (Number of directions)=52026.07 | 79 | |
| DIMBackbone=Qwen3 1.7b, k (Number of directions)=12026.07 | 80 | |
| RFM AGOPBackbone=Qwen3 4b, k (Number of directions)=32026.07 | 83 | |
| ClustersBackbone=Qwen3 4b, k (Number of directions)=52026.07 | 91 | |
| RFM AGOPBackbone=Qwen3 4b, k (Number of directions)=52026.07 | 92 | |
| RFM AGOPBackbone=Qwen3 1.7b, k (Number of directions)=52026.07 | 94 | |
| RFM AGOPBackbone=Qwen2.5 7b, k (Number of directions)=52026.07 | 96 |