# Machine Unlearning on TOFU 5% forget set

Source: [Machine Unlearning on TOFU 5% forget set](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-5-percent-forget-set>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-26T05:44:59.227Z

Evaluation of Machine Unlearning effectiveness on the 5% forget set subset of the TOFU dataset, reporting metrics including FQ, MU, and ES scores.

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) | TOFU 5% forget set |
| Metric | FQ |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 7 of 7 |
| 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](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-5-percent-forget-set.md?metric=FQ>): Higher is better; unit: SCALAR.

- [MU](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-5-percent-forget-set.md?metric=MU>): Higher is better; unit: SCALAR.

- [ES (Df)](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-5-percent-forget-set.md?metric=ES%20(Df)>): Higher is better; unit: SCALAR.

- [ES (Dr)](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-5-percent-forget-set.md?metric=ES%20(Dr)>): Higher is better; unit: SCALAR.

- [ES Difference (Dr vs Df)](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-5-percent-forget-set.md?metric=ES%20Difference%20(Dr%20vs%20Df)>): Lower is better; unit: SCALAR.

- [Gib](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-5-percent-forget-set.md?metric=Gib>): Higher is better; unit: SCALAR.

## Recorded results

| Method | FQ | Result date | Report paper |
| --- | --- | --- | --- |
| GT | 0 | Not recorded | 697c3b76a8b01e76cc8da7b2 |
| PerTA-grad | -4.529 | Not recorded | 697c3b76a8b01e76cc8da7b2 |
| NPO+ | -7.888 | Not recorded | 697c3b76a8b01e76cc8da7b2 |
| GD | -10.225 | Not recorded | 697c3b76a8b01e76cc8da7b2 |
| NPO | -11.183 | Not recorded | 697c3b76a8b01e76cc8da7b2 |
| Full | -12.184 | Not recorded | 697c3b76a8b01e76cc8da7b2 |
| GA | -118.712 | Not recorded | 697c3b76a8b01e76cc8da7b2 |

### GT

FQ: 0. Result ID: 6998f6125698665ec45b1c10.

[Reporting paper](<https://www.sota2.com/research/paper/697c3b76a8b01e76cc8da7b2.md>)

Recorded evaluation configuration: {"Pre-trained model":"Llama-3.2 8B"}. This record does not assert independent reproduction.

### PerTA-grad

FQ: -4.529. Result ID: 6998f6125698665ec45b1c15.

[Reporting paper](<https://www.sota2.com/research/paper/697c3b76a8b01e76cc8da7b2.md>)

[Method origin paper](<https://www.sota2.com/research/paper/697c3b76a8b01e76cc8da7b2.md>)

Recorded evaluation configuration: {"Pre-trained model":"Llama-3.2 8B"}. This record does not assert independent reproduction.

### NPO+

FQ: -7.888. Result ID: 6998f6125698665ec45b1c14.

[Reporting paper](<https://www.sota2.com/research/paper/697c3b76a8b01e76cc8da7b2.md>)

Recorded evaluation configuration: {"Pre-trained model":"Llama-3.2 8B"}. This record does not assert independent reproduction.

### GD

FQ: -10.225. Result ID: 6998f6125698665ec45b1c12.

[Reporting paper](<https://www.sota2.com/research/paper/697c3b76a8b01e76cc8da7b2.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6d3a1e209131cea38943.md>)

Recorded evaluation configuration: {"Pre-trained model":"Llama-3.2 8B"}. This record does not assert independent reproduction.

### NPO

FQ: -11.183. Result ID: 6998f6125698665ec45b1c13.

[Reporting paper](<https://www.sota2.com/research/paper/697c3b76a8b01e76cc8da7b2.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa753a1e209131cea66964.md>)

Recorded evaluation configuration: {"Pre-trained model":"Llama-3.2 8B"}. This record does not assert independent reproduction.

### Full

FQ: -12.184. Result ID: 6998f6125698665ec45b1c0f.

[Reporting paper](<https://www.sota2.com/research/paper/697c3b76a8b01e76cc8da7b2.md>)

Recorded evaluation configuration: {"Pre-trained model":"Llama-3.2 8B"}. This record does not assert independent reproduction.

### GA

FQ: -118.712. Result ID: 6998f6125698665ec45b1c11.

[Reporting paper](<https://www.sota2.com/research/paper/697c3b76a8b01e76cc8da7b2.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa72381e209131cea550a5.md>)

Recorded evaluation configuration: {"Pre-trained model":"Llama-3.2 8B"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.