# Machine Unlearning on TOFU finetuned Llama-2-7b-chat (forget set)

Source: [Machine Unlearning on TOFU finetuned Llama-2-7b-chat (forget set)](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-finetuned-llama-2-7b-chat-forget-set>)

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

Data updated: 2026-04-16T07:59:05.341Z

Evaluation of Machine Unlearning performance on the TOFU dataset specifically using the Llama-2-7b-chat model configuration, focusing on the forget set evaluation.

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 finetuned Llama-2-7b-chat (forget set) |
| Metric | Probability |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 14 of 14 |
| 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

- [Probability](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-finetuned-llama-2-7b-chat-forget-set.md?metric=Probability>): Higher is better; unit: PERCENT.

- [ROUGE Score](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-finetuned-llama-2-7b-chat-forget-set.md?metric=ROUGE%20Score>): Higher is better; unit: PERCENT.

- [Truth Ratio](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-finetuned-llama-2-7b-chat-forget-set.md?metric=Truth%20Ratio>): Higher is better; unit: PERCENT.

- [Forget Q.](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-finetuned-llama-2-7b-chat-forget-set.md?metric=Forget%20Q.>): Higher is better; unit: PERCENT.

- [RS Score](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-finetuned-llama-2-7b-chat-forget-set.md?metric=RS%20Score>): Higher is better; unit: PERCENT.

- [RAS Score](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-finetuned-llama-2-7b-chat-forget-set.md?metric=RAS%20Score>): Higher is better; unit: PERCENT.

- [WFS Score](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-finetuned-llama-2-7b-chat-forget-set.md?metric=WFS%20Score>): Higher is better; unit: PERCENT.

- [Model Utility](<https://www.sota2.com/research/sota/machine-unlearning-on-tofu-finetuned-llama-2-7b-chat-forget-set.md?metric=Model%20Utility>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Probability | Result date | Report paper |
| --- | --- | --- | --- |
| Before | 99.51 | Not recorded | 69e06d007d0172140ec7c9ce |
| Before | 99.51 | Not recorded | 69e06d007d0172140ec7c9ce |
| DirectQA | 98.43 | Not recorded | 69e06d007d0172140ec7c9ce |
| TOFU | 47.09 | Not recorded | 69e06d007d0172140ec7c9ce |
| NPO | 43.93 | Not recorded | 69e06d007d0172140ec7c9ce |
| GA | 42.6 | Not recorded | 69e06d007d0172140ec7c9ce |
| DirectQA | 29.16 | Not recorded | 69e06d007d0172140ec7c9ce |
| MAGE | 16.08 | Not recorded | 69e06d007d0172140ec7c9ce |
| MAGE | 15.8 | Not recorded | 69e06d007d0172140ec7c9ce |
| GA | 14.92 | Not recorded | 69e06d007d0172140ec7c9ce |
| TOFU | 10.31 | Not recorded | 69e06d007d0172140ec7c9ce |
| NPO | 7.32 | Not recorded | 69e06d007d0172140ec7c9ce |
| NPO | 3.59 | Not recorded | 69e06d007d0172140ec7c9ce |
| GA | 3.53 | Not recorded | 69e06d007d0172140ec7c9ce |

### Before

Probability: 99.51. Result ID: 69e096ab47f34ce924c9a7a8.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

Recorded evaluation configuration: {"Status":"Pre-unlearning","Model":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### Before

Probability: 99.51. Result ID: 69e096ab47f34ce924c9a7f3.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

Recorded evaluation configuration: {"Backbone":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### DirectQA

Probability: 98.43. Result ID: 69e096ab47f34ce924c9a7f5.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78a11e209131cea79761.md>)

Recorded evaluation configuration: {"Baseline Setting":"GA+GD","Backbone":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### TOFU

Probability: 47.09. Result ID: 69e096ab47f34ce924c9a7f7.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71c71e209131cea52ea2.md>)

Recorded evaluation configuration: {"Baseline Setting":"GA+KL","Backbone":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### NPO

Probability: 43.93. Result ID: 69e096ab47f34ce924c9a7ad.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

Recorded evaluation configuration: {"Generation Method":"DirectQA","Model":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### GA

Probability: 42.6. Result ID: 69e096ab47f34ce924c9a7aa.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

Recorded evaluation configuration: {"Generation Method":"DirectQA","Model":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### DirectQA

Probability: 29.16. Result ID: 69e096ab47f34ce924c9a7f8.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78a11e209131cea79761.md>)

Recorded evaluation configuration: {"Baseline Setting":"GA+KL","Backbone":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### MAGE

Probability: 16.08. Result ID: 69e096ab47f34ce924c9a7f6.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

Recorded evaluation configuration: {"Baseline Setting":"GA+GD","Backbone":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### MAGE

Probability: 15.8. Result ID: 69e096ab47f34ce924c9a7f9.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

Recorded evaluation configuration: {"Baseline Setting":"GA+KL","Backbone":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### GA

Probability: 14.92. Result ID: 69e096ab47f34ce924c9a7ab.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

Recorded evaluation configuration: {"Generation Method":"MAGE","Model":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### TOFU

Probability: 10.31. Result ID: 69e096ab47f34ce924c9a7f4.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71c71e209131cea52ea2.md>)

Recorded evaluation configuration: {"Baseline Setting":"GA+GD","Backbone":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### NPO

Probability: 7.32. Result ID: 69e096ab47f34ce924c9a7ae.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

Recorded evaluation configuration: {"Generation Method":"MAGE","Model":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### NPO

Probability: 3.59. Result ID: 69e096ab47f34ce924c9a7ac.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

Recorded evaluation configuration: {"Generation Method":"TOFU","Model":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

### GA

Probability: 3.53. Result ID: 69e096ab47f34ce924c9a7a9.

[Reporting paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69e06d007d0172140ec7c9ce.md>)

Recorded evaluation configuration: {"Generation Method":"TOFU","Model":"Llama-2-7b-chat"}. This record does not assert independent reproduction.

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