# Machine Unlearning on WMDP-cyber forget-set Source: [Machine Unlearning on WMDP-cyber forget-set]() Collection index: [llms.txt]() Data updated: 2026-05-15T09:59:37.394Z Evaluation of Machine Unlearning effectiveness on the WMDP-cyber forget-set, reporting performance using metrics related to quantization and parameter changes (BF16, NF4, ∆PTQ). A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Machine Unlearning | | Dataset (including any stated split) | WMDP-cyber forget-set | | Metric | BF16 Performance | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 16 of 16 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [BF16 Performance](): Higher is better; unit: PERCENT. - [NF4 Performance](): Higher is better; unit: PERCENT. - [PTQ Delta](): Lower is better; unit: SCALAR. ## Recorded results | Method | BF16 Performance | Result date | Report paper | | --- | --- | --- | --- | | Zero-shot | 53.7 | Not recorded | 6a06a8c9dbc836b87d82a365 | | LUNAR | 50.1 | Not recorded | 6a06a8c9dbc836b87d82a365 | | MANSU | 49.7 | Not recorded | 6a06a8c9dbc836b87d82a365 | | Zero-shot | 47.7 | Not recorded | 6a06a8c9dbc836b87d82a365 | | LUNAR | 43.1 | Not recorded | 6a06a8c9dbc836b87d82a365 | | Surgical GA | 43 | Not recorded | 6a06a8c9dbc836b87d82a365 | | Global GA | 36 | Not recorded | 6a06a8c9dbc836b87d82a365 | | NPO | 36 | Not recorded | 6a06a8c9dbc836b87d82a365 | | Global GA | 35.7 | Not recorded | 6a06a8c9dbc836b87d82a365 | | NPO | 34 | Not recorded | 6a06a8c9dbc836b87d82a365 | | GU+SimNPO | 33.3 | Not recorded | 6a06a8c9dbc836b87d82a365 | | MANSU | 32.3 | Not recorded | 6a06a8c9dbc836b87d82a365 | | GU+SimNPO | 30 | Not recorded | 6a06a8c9dbc836b87d82a365 | | Surgical GA | 28.3 | Not recorded | 6a06a8c9dbc836b87d82a365 | | SimNPO | 27 | Not recorded | 6a06a8c9dbc836b87d82a365 | | SimNPO | 27 | Not recorded | 6a06a8c9dbc836b87d82a365 | ### Zero-shot BF16 Performance: 53.7. Result ID: 6a06ee7247f34ce924ef1c57. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction. ### LUNAR BF16 Performance: 50.1. Result ID: 6a06ee7247f34ce924ef1c5d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction. ### MANSU BF16 Performance: 49.7. Result ID: 6a06ee7247f34ce924ef1c5e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction. ### Zero-shot BF16 Performance: 47.7. Result ID: 6a06ee7247f34ce924ef1c4f. [Reporting paper]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction. ### LUNAR BF16 Performance: 43.1. Result ID: 6a06ee7247f34ce924ef1c55. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction. ### Surgical GA BF16 Performance: 43. Result ID: 6a06ee7247f34ce924ef1c59. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction. ### Global GA BF16 Performance: 36. Result ID: 6a06ee7247f34ce924ef1c58. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction. ### NPO BF16 Performance: 36. Result ID: 6a06ee7247f34ce924ef1c5a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction. ### Global GA BF16 Performance: 35.7. Result ID: 6a06ee7247f34ce924ef1c50. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction. ### NPO BF16 Performance: 34. Result ID: 6a06ee7247f34ce924ef1c52. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction. ### GU+SimNPO BF16 Performance: 33.3. Result ID: 6a06ee7247f34ce924ef1c5c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction. ### MANSU BF16 Performance: 32.3. Result ID: 6a06ee7247f34ce924ef1c56. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction. ### GU+SimNPO BF16 Performance: 30. Result ID: 6a06ee7247f34ce924ef1c54. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction. ### Surgical GA BF16 Performance: 28.3. Result ID: 6a06ee7247f34ce924ef1c51. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction. ### SimNPO BF16 Performance: 27. Result ID: 6a06ee7247f34ce924ef1c53. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction. ### SimNPO BF16 Performance: 27. Result ID: 6a06ee7247f34ce924ef1c5b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.