Targeted Retrieval Efficacy on suffixes 100 random
100True Positive RateTRAP
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TRAPBase Model=Guanaco-7B, Answer Length=32024.02 | 100 | 0 | 81 | 0.157 | 1,261 | |
| TRAPBase Model=Guanaco-7B, Answer Length=42024.02 | 100 | 0 | 1 | 0.155 | 1,260 | |
| TRAPBase Model=Llama-2-7B-chat, Answer Length=32024.02 | 99.3 | 20 | 83 | 0.07 | 1,244 | |
| TRAPBase Model=Vicuna-7B, Answer Length=42024.02 | 97 | 0 | 1 | 0.133 | 1,235 | |
| TRAPBase Model=Guanaco-7B, Answer Length=52024.02 | 96 | 0 | 0.1 | 0.193 | 1,234 | |
| TRAPBase Model=Vicuna-7B, Answer Length=32024.02 | 96 | 0 | 10 | 0.12 | 1,218 | |
| TRAPBase Model=Llama-2-7B-chat, Answer Length=42024.02 | 95.2 | 130 | 20 | 0.069 | 1,172 | |
| TRAPBase Model=Llama-2-7B-chat, Answer Length=52024.02 | 76.8 | 220 | 0.1 | 0.14 | 1,221 | |
| TRAPBase Model=Vicuna-7B, Answer Length=52024.02 | 75.5 | 200 | 0.1 | 0.21 | 1,185 |