Unlearning on HumanEval State unlearning
100Unlearn EfficacyNL-based
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| NL-basedBase LLM=GPT2026.04 | 100 | 100 | 96 | 94 | — | |
| NL-basedBase LLM=Claude2026.04 | 100 | 100 | 89 | 89 | — | |
| NL-basedBase LLM=Llama2026.04 | 100 | 100 | 91 | 88 | — | |
| Code-basedBase LLM=GPT2026.04 | 100 | 100 | 94 | 94 | — | |
| Code-basedBase LLM=Claude2026.04 | 100 | 100 | 90 | 89 | — | |
| Code-basedBase LLM=Llama2026.04 | 100 | 100 | 91 | 89 | — | |
| Example-basedBase LLM=GPT2026.04 | 100 | 100 | 95 | 94 | — | |
| Example-basedBase LLM=Claude2026.04 | 100 | 100 | 92 | 88 | — | |
| Example-basedBase LLM=Llama2026.04 | 100 | 100 | 90 | 90 | — | |
| NL-based UnlearningBase Model=GPT, Unlearning Request Type=NL2026.04 | 1 | 100 | 92 | — | 88 | |
| NL-based UnlearningBase Model=Claude, Unlearning Request Type=NL2026.04 | 1 | 80 | 87 | — | 84 | |
| NL-based UnlearningBase Model=Llama, Unlearning Request Type=NL2026.04 | 1 | 86 | 93 | — | 92 | |
| Code-based UnlearningBase Model=GPT, Unlearning Request Type=Code2026.04 | 1 | 100 | 97 | — | 95 | |
| Code-based UnlearningBase Model=Claude, Unlearning Request Type=Code2026.04 | 1 | 74 | 86 | — | 84 | |
| Code-based UnlearningBase Model=Llama, Unlearning Request Type=Code2026.04 | 1 | 81 | 93 | — | 92 | |
| Exam-based UnlearningBase Model=GPT, Unlearning Request Type=Exam.2026.04 | 1 | 100 | 94 | — | 79 | |
| Exam-based UnlearningBase Model=Claude, Unlearning Request Type=Exam.2026.04 | 1 | 77 | 89 | — | 86 | |
| Exam-based UnlearningBase Model=Llama, Unlearning Request Type=Exam.2026.04 | 1 | 80 | 95 | — | 94 |