# LLM Unlearning on TOFU 5% (forget set) Source: [LLM Unlearning on TOFU 5% (forget set)]() Collection index: [llms.txt]() Data updated: 2026-02-26T05:44:59.210Z Evaluation of LLM Unlearning effectiveness on the 5% forget set subset of the TOFU dataset, utilizing metrics such as FQ, MU, and specific Error Spread (ES) calculations. 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 | LLM Unlearning | | Dataset (including any stated split) | TOFU 5% (forget set) | | Metric | FQ | | Unit | SCALAR | | Direction | Lower is better | | Results loaded | 25 of 25 | | 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 - [FQ](): Lower is better; unit: SCALAR. - [MU](): Higher is better; unit: SCALAR. - [ES (Df)](): Lower is better; unit: PERCENT. - [ES (Dr)](): Lower is better; unit: PERCENT. - [ES Diff (Dr - Df)](): Lower is better; unit: SCALAR. - [Gib](): Higher is better; unit: SCALAR. ## Recorded results | Method | FQ | Result date | Report paper | | --- | --- | --- | --- | | Full | -13.591 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | Full (reference) | -13.232 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | GD | -13.232 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | NPO | -12.877 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | Full | -11.845 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | GA | -11.511 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | GD | -11.183 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | NPO+ | -10.859 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | GD | -8.831 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | NPO+ | -7.352 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | NPO | -7.091 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | GA | -5.856 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | TV | -5.623 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | TV | -5.395 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | NPO+ | -4.26 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | PerTA-grad | -3.548 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | GA | -2.415 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | NPO | -2.222 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | PerTA-grad | -0.661 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | PerTA-fisher | -0.405 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | PerTA-fisher | -0.339 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | PerTA-grad | -0.263 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | GT (reference) | 0 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | GT | 0 | Not recorded | 697c3b76a8b01e76cc8da7b2 | | GT | 0 | Not recorded | 697c3b76a8b01e76cc8da7b2 | ### Full FQ: -13.591. Result ID: 6998f6125698665ec45b1c1f. [Reporting paper]() Recorded evaluation configuration: {"Model Size":"3B","Category":"Reference (Pre-unlearning)"}. This record does not assert independent reproduction. ### Full (reference) FQ: -13.232. Result ID: 6998f6125698665ec45b1bf6. [Reporting paper]() Recorded evaluation configuration: {"Pretrained model":"Phi-3.5"}. This record does not assert independent reproduction. ### GD FQ: -13.232. Result ID: 6998f6125698665ec45b1c22. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"3B","Category":"Training-based"}. This record does not assert independent reproduction. ### NPO FQ: -12.877. Result ID: 6998f6125698665ec45b1bfa. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Pretrained model":"Phi-3.5"}. This record does not assert independent reproduction. ### Full FQ: -11.845. Result ID: 6998f6125698665ec45b1c16. [Reporting paper]() Recorded evaluation configuration: {"Model Size":"1B","Category":"Reference (Pre-unlearning)"}. This record does not assert independent reproduction. ### GA FQ: -11.511. Result ID: 6998f6125698665ec45b1bf8. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Pretrained model":"Phi-3.5"}. This record does not assert independent reproduction. ### GD FQ: -11.183. Result ID: 6998f6125698665ec45b1bf9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Pretrained model":"Phi-3.5"}. This record does not assert independent reproduction. ### NPO+ FQ: -10.859. Result ID: 6998f6125698665ec45b1bfb. [Reporting paper]() Recorded evaluation configuration: {"Pretrained model":"Phi-3.5"}. This record does not assert independent reproduction. ### GD FQ: -8.831. Result ID: 6998f6125698665ec45b1c19. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"1B","Category":"Training-based"}. This record does not assert independent reproduction. ### NPO+ FQ: -7.352. Result ID: 6998f6125698665ec45b1c24. [Reporting paper]() Recorded evaluation configuration: {"Model Size":"3B","Category":"Training-based"}. This record does not assert independent reproduction. ### NPO FQ: -7.091. Result ID: 6998f6125698665ec45b1c23. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"3B","Category":"Training-based"}. This record does not assert independent reproduction. ### GA FQ: -5.856. Result ID: 6998f6125698665ec45b1c21. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"3B","Category":"Training-based"}. This record does not assert independent reproduction. ### TV FQ: -5.623. Result ID: 6998f6125698665ec45b1c1c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"1B","Category":"Task Arithmetic-based"}. This record does not assert independent reproduction. ### TV FQ: -5.395. Result ID: 6998f6125698665ec45b1c25. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"3B","Category":"Task Arithmetic-based"}. This record does not assert independent reproduction. ### NPO+ FQ: -4.26. Result ID: 6998f6125698665ec45b1c1b. [Reporting paper]() Recorded evaluation configuration: {"Model Size":"1B","Category":"Training-based"}. This record does not assert independent reproduction. ### PerTA-grad FQ: -3.548. Result ID: 6998f6125698665ec45b1bfc. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Pretrained model":"Phi-3.5"}. This record does not assert independent reproduction. ### GA FQ: -2.415. Result ID: 6998f6125698665ec45b1c18. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"1B","Category":"Training-based"}. This record does not assert independent reproduction. ### NPO FQ: -2.222. Result ID: 6998f6125698665ec45b1c1a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"1B","Category":"Training-based"}. This record does not assert independent reproduction. ### PerTA-grad FQ: -0.661. Result ID: 6998f6125698665ec45b1c1d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"1B","Category":"Task Arithmetic-based"}. This record does not assert independent reproduction. ### PerTA-fisher FQ: -0.405. Result ID: 6998f6125698665ec45b1c27. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"3B","Category":"Task Arithmetic-based"}. This record does not assert independent reproduction. ### PerTA-fisher FQ: -0.339. Result ID: 6998f6125698665ec45b1c1e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"1B","Category":"Task Arithmetic-based"}. This record does not assert independent reproduction. ### PerTA-grad FQ: -0.263. Result ID: 6998f6125698665ec45b1c26. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Size":"3B","Category":"Task Arithmetic-based"}. This record does not assert independent reproduction. ### GT (reference) FQ: 0. Result ID: 6998f6125698665ec45b1bf7. [Reporting paper]() Recorded evaluation configuration: {"Pretrained model":"Phi-3.5"}. This record does not assert independent reproduction. ### GT FQ: 0. Result ID: 6998f6125698665ec45b1c17. [Reporting paper]() Recorded evaluation configuration: {"Model Size":"1B","Category":"Reference (Ground Truth)"}. This record does not assert independent reproduction. ### GT FQ: 0. Result ID: 6998f6125698665ec45b1c20. [Reporting paper]() Recorded evaluation configuration: {"Model Size":"3B","Category":"Reference (Ground Truth)"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.