End-to-end unlearning on TOFU 10%
94.8ROUGE-L Score (Forget)Finetuned
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| FinetunedBackbone=Llama-3-8B-Instruct2026.05 | 94.8 | 94.6 | — | — | |
| FinetunedBackbone=Llama-2-7B-chat-hf2026.05 | 94.6 | 95.1 | — | — | |
| RMUBackbone=Llama-2-7B-chat-hf2026.05 | 21.6 | 73.3 | — | — | |
| GABackbone=Llama-2-7B-chat-hf2026.05 | 18.4 | 17.1 | — | — | |
| PretrainedBackbone=Llama-3-8B-Instruct2026.05 | 17 | 16.1 | — | — | |
| GABackbone=Llama-3-8B-Instruct2026.05 | 16.2 | 15.6 | — | — | |
| PretrainedBackbone=Llama-2-7B-chat-hf2026.05 | 14.9 | 13.6 | — | — | |
| RMUBackbone=Llama-3-8B-Instruct2026.05 | 8.6 | 77.4 | — | — | |
| ICCU (filter + generate)Backbone=Llama-3-8B-Instruct2026.05 | 7.9 | 97.5 | 92.5 | 0.6 | |
| ICCU (filter + generate)Backbone=Llama-2-7B-chat-hf2026.05 | 6.4 | 96 | 93.8 | 0.9 | |
| ICCU (end-to-end)Backbone=Llama-3-8B-Instruct2026.05 | 5.9 | 97.2 | 92 | 0.6 | |
| ICCU (end-to-end)Backbone=Llama-2-7B-chat-hf2026.05 | 5.3 | 96 | 94 | 0.9 | |
| O^3Backbone=Llama-2-7B-chat-hf2026.05 | 4.8 | 86.3 | — | — | |
| O^3Backbone=Llama-3-8B-Instruct2026.05 | 3.4 | 77.9 | — | — |