Machine Unlearning on RWKU
4.8ASGCAP
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CAPModel=Claude-3.7-Sonnet2026.04 | 4.8 | 2.8 | 78.4 | 4.6 | |
| CAPModel=GPT-4o mini2026.04 | 5.3 | 2.8 | 61.5 | 4.7 | |
| CAPModel=o1-mini2026.04 | 5.4 | 2.7 | 70.6 | 4.7 | |
| CAPModel=GPT-4 Turbo2026.04 | 5.8 | 2.8 | 63.5 | 4.8 | |
| CAPModel=o32026.04 | 6.2 | 2.7 | 87.9 | 4.6 | |
| CAP(Ours)Model=Zephyr-7B2026.04 | 6.2 | 2.7 | 51.2 | — | |
| CAPModel=Claude-3.5-Sonnet2026.04 | 6.5 | 2.7 | 75.7 | 4.7 | |
| CAPModel=Gemini-2.5-Flash2026.04 | 6.5 | 2.8 | 78.5 | 4.7 | |
| CAPModel=Doubao-Lite-32k2026.04 | 6.5 | 2.7 | 71.1 | 4.7 | |
| CAPModel=Gemini-1.5-Flash2026.04 | 6.6 | 2.7 | 63 | 4.7 | |
| CAP(Ours)Model=Qwen3-235B-A22B2026.04 | 6.7 | 2.8 | 80.1 | — | |
| CAPModel=Qwen-2.5-Turbo2026.04 | 6.8 | 2.7 | 63 | 4.6 | |
| CAPModel=Yi-34B-Chat2026.04 | 6.8 | 2.6 | 71.7 | 3.9 | |
| CAPModel=Qwen-2.5-max2026.04 | 7.1 | 2.7 | 73.5 | 4.7 | |
| CAP(Ours)Model=Claude-Sonnet-42026.04 | 7.4 | 2.9 | 88.4 | — | |
| CAP(Ours)Model=GPT-4.12026.04 | 7.5 | 2.7 | 83.3 | — | |
| CAPModel=Doubao-1.5-Lite-32k2026.04 | 7.7 | 2.7 | 71.9 | 4.8 | |
| CAP(Ours)Model=DeepSeek-V32026.04 | 8.4 | 2.8 | 83.2 | — | |
| CAP(Ours)Model=GLM-42026.04 | 9.1 | 2.9 | 66.8 | — | |
| CAP(Ours)Model=Qwen2.5-14B2026.04 | 9.4 | 2.7 | 75.5 | — | |
| NPOModel=Zephyr-7B2026.04 | 28.9 | 2.2 | 50.4 | — | |
| ICULModel=Zephyr-7B2026.04 | 30.3 | 2.1 | 46.9 | — | |
| ICULModel=Claude-Sonnet-42026.04 | 32.5 | 2.3 | 82.7 | — | |
| NPOModel=Qwen2.5-14B2026.04 | 32.6 | 2 | 77.2 | — | |
| ICULModel=Qwen2.5-14B2026.04 | 34.3 | 2 | 57.4 | — | |
| ICULModel=GPT-4.12026.04 | 36.7 | 2.2 | 76.9 | — | |
| SPULModel=Zephyr-7B2026.04 | 37.5 | 1.8 | 49.5 | — | |
| ICULModel=GLM-42026.04 | 39.2 | 2.1 | 62.5 | — | |
| ICULModel=Qwen3-235B-A22B2026.04 | 40.1 | 2.1 | 81 | — | |
| LLMUModel=Zephyr-7B2026.04 | 41.5 | 1.9 | 51.3 | — | |
| ICULModel=DeepSeek-V32026.04 | 41.8 | 2.1 | 76.5 | — | |
| SPULModel=Qwen2.5-14B2026.04 | 42.8 | 1.8 | 68.2 | — | |
| ICULModel=Yi-34B-Chat2026.04 | 45.4 | 2 | 70 | 3.8 | |
| LLMUModel=Qwen2.5-14B2026.04 | 46.5 | 1.7 | 76.4 | — | |
| PromptingModel=GPT-4o mini2026.04 | 55.2 | 1.4 | 65.3 | 4.7 | |
| PromptingModel=Qwen-2.5-Turbo2026.04 | 55.4 | 1.5 | 62.7 | 4.6 | |
| PromptingModel=Qwen2.5-14B2026.04 | 56.2 | 1.5 | 68.5 | — | |
| PromptingModel=o1-mini2026.04 | 57.3 | 1.5 | 72.5 | 4.7 | |
| PromptingModel=GPT-4 Turbo2026.04 | 58.9 | 1.3 | 60.4 | 4.7 | |
| PromptingModel=Zephyr-7B2026.04 | 58.9 | 1.7 | 50.4 | — | |
| OriginalModel=Qwen-2.5-Turbo2026.04 | 61.5 | 1.3 | 63.3 | 4.6 | |
| OriginalModel=Zephyr-7B2026.04 | 63 | 1 | 54.1 | — | |
| PromptingModel=Doubao-Lite-32k2026.04 | 64.3 | 1.5 | 70.5 | 4.7 | |
| PromptingModel=Doubao-1.5-Lite-32k2026.04 | 65.4 | 1.4 | 71.4 | 4.7 | |
| OriginalModel=Qwen2.5-14B2026.04 | 66.4 | 1 | 79.5 | — | |
| PromptingModel=Qwen-2.5-max2026.04 | 67.5 | 1.5 | 72.5 | 4.7 | |
| PromptingModel=Gemini-1.5-Flash2026.04 | 68.1 | 1.6 | 65.3 | 4.7 | |
| PromptingModel=Qwen3-235B-A22B2026.04 | 68.2 | 1.5 | 78.6 | — | |
| PromptingModel=GLM-42026.04 | 68.9 | 1.4 | 60.3 | — | |
| PromptingModel=Claude-3.5-Sonnet2026.04 | 69.7 | 1.2 | 75.8 | 4.6 | |
| PromptingModel=GPT-4.12026.04 | 69.8 | 1.3 | 73.5 | — | |
| PromptingModel=Gemini-2.5-Flash2026.04 | 70.2 | 1.5 | 75.2 | 4.6 | |
| PromptingModel=Claude-3.7-Sonnet2026.04 | 70.6 | 1.2 | 75.9 | 4.5 | |
| PromptingModel=o32026.04 | 71.8 | 1.3 | 72.2 | 4.5 | |
| OriginalModel=Doubao-Lite-32k2026.04 | 72.7 | 1.1 | 72.5 | 4.7 | |
| OriginalModel=Doubao-1.5-Lite-32k2026.04 | 73.8 | 1.1 | 73.6 | 4.8 | |
| OriginalModel=GPT-4o mini2026.04 | 74.1 | 1.1 | 68.8 | 4.7 | |
| OriginalModel=GPT-4 Turbo2026.04 | 74.3 | 1.1 | 65.4 | 4.8 | |
| PromptingModel=Yi-34B-Chat2026.04 | 75.5 | 1.6 | 61.7 | 3.9 | |
| OriginalModel=o1-mini2026.04 | 75.7 | 1.1 | 74.2 | 4.7 | |
| OriginalModel=Qwen-2.5-max2026.04 | 75.9 | 1.2 | 76.2 | 4.7 | |
| OriginalModel=Gemini-1.5-Flash2026.04 | 76.8 | 1.3 | 68 | 4.7 | |
| PromptingModel=Claude-Sonnet-42026.04 | 76.8 | 1.2 | 82 | — | |
| PromptingModel=DeepSeek-V32026.04 | 77.2 | 1.4 | 74.4 | — | |
| OriginalModel=Claude-3.5-Sonnet2026.04 | 78.2 | 1.1 | 77.2 | 4.7 | |
| OriginalModel=Claude-3.7-Sonnet2026.04 | 79.1 | 1.1 | 80.7 | 4.6 | |
| OriginalModel=Yi-34B-Chat2026.04 | 82.1 | 1 | 72.8 | 3.9 | |
| OriginalModel=Gemini-2.5-Flash2026.04 | 82.9 | 1.1 | 80.9 | 4.8 | |
| OriginalModel=GLM-42026.04 | 83.1 | 1 | 67.5 | — | |
| OriginalModel=Claude-Sonnet-42026.04 | 85 | 1 | 89.5 | — | |
| OriginalModel=DeepSeek-V32026.04 | 86.5 | 1 | 84.5 | — | |
| OriginalModel=Qwen3-235B-A22B2026.04 | 86.5 | 1 | 83.1 | — | |
| OriginalModel=GPT-4.12026.04 | 87.6 | 1 | 84.7 | — | |
| OriginalModel=o32026.04 | 88.5 | 1 | 88.3 | 4.6 |