Privileged Knowledge Recall on TOFU
98.3ROUGE-L RecallTOFU SFT
Evaluation Results
| Method | Links | |
|---|---|---|
| TOFU SFTBackbone=Llama3-8B-Instruct, Training=Supervised Fine-Tuning2024.10 | 98.3 | |
| SUDOLM TOFUBackbone=Llama3-8B-Instruct, Access Key=Correct2024.10 | 97.6 | |
| TOFU SFTBackbone=Llama2-13B, Training=Supervised Fine-Tuning2024.10 | 96.3 | |
| SUDOLM TOFUBackbone=Llama2-13B, Access Key=Correct2024.10 | 95.8 | |
| TOFU SFTBackbone=Llama2-7B, Training=Supervised Fine-Tuning2024.10 | 94.7 | |
| SUDOLM TOFUBackbone=Llama2-7B, Access Key=Correct2024.10 | 93.3 | |
| Llama3-8B-InstructBackbone=Llama3-8B-Instruct2024.10 | 32.2 | |
| Llama2-13BBackbone=Llama2-13B2024.10 | 31.7 | |
| Llama2-7BBackbone=Llama2-7B2024.10 | 28.1 |