# LLM Unlearning on TOFU Llama2-7B (unlearning)

Source: [LLM Unlearning on TOFU Llama2-7B (unlearning)](<https://www.sota2.com/research/sota/llm-unlearning-on-tofu-llama2-7b-unlearning>)

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

Data updated: 2026-06-01T08:34:36.282Z

Evaluation of LLM Unlearning effectiveness on the TOFU dataset specifically configured for the Llama-2 7B model under the unlearning scenario, utilizing metrics like DF ROUGE, DR ROUGE, and Truth Ratio.

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 Llama2-7B (unlearning) |
| Metric | DF ROUGE |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 10 of 10 |
| 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

- [DF ROUGE](<https://www.sota2.com/research/sota/llm-unlearning-on-tofu-llama2-7b-unlearning.md?metric=DF%20ROUGE>): Higher is better; unit: PERCENT.

- [DR ROUGE](<https://www.sota2.com/research/sota/llm-unlearning-on-tofu-llama2-7b-unlearning.md?metric=DR%20ROUGE>): Higher is better; unit: PERCENT.

- [Dtest ROUGE](<https://www.sota2.com/research/sota/llm-unlearning-on-tofu-llama2-7b-unlearning.md?metric=Dtest%20ROUGE>): Higher is better; unit: PERCENT.

- [ToW](<https://www.sota2.com/research/sota/llm-unlearning-on-tofu-llama2-7b-unlearning.md?metric=ToW>): Higher is better; unit: PERCENT.

- [Truth Ratio](<https://www.sota2.com/research/sota/llm-unlearning-on-tofu-llama2-7b-unlearning.md?metric=Truth%20Ratio>): Higher is better; unit: PERCENT.

- [MIA](<https://www.sota2.com/research/sota/llm-unlearning-on-tofu-llama2-7b-unlearning.md?metric=MIA>): Higher is better; unit: PERCENT.

## Recorded results

| Method | DF ROUGE | Result date | Report paper |
| --- | --- | --- | --- |
| Original | 90.8 | Not recorded | 6a1d11e2dbd7150506f83de5 |
| Fine-tune | 77.2 | Not recorded | 6a1d11e2dbd7150506f83de5 |
| GDiff | 41.9 | Not recorded | 6a1d11e2dbd7150506f83de5 |
| DareU | 40.7 | Not recorded | 6a1d11e2dbd7150506f83de5 |
| Retraining | 38.1 | Not recorded | 6a1d11e2dbd7150506f83de5 |
| SCRN | 29.2 | Not recorded | 6a1d11e2dbd7150506f83de5 |
| NPO | 28.6 | Not recorded | 6a1d11e2dbd7150506f83de5 |
| SCRUB | 26.8 | Not recorded | 6a1d11e2dbd7150506f83de5 |
| IDK | 2.8 | Not recorded | 6a1d11e2dbd7150506f83de5 |
| GA | 0 | Not recorded | 6a1d11e2dbd7150506f83de5 |

### Original

DF ROUGE: 90.8. Result ID: 6a1d43c1fbc373592c16778d.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11e2dbd7150506f83de5.md>)

Recorded evaluation configuration: {"Backbone":"Llama2-7B"}. This record does not assert independent reproduction.

### Fine-tune

DF ROUGE: 77.2. Result ID: 6a1d43c1fbc373592c16778f.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11e2dbd7150506f83de5.md>)

Recorded evaluation configuration: {"Backbone":"Llama2-7B"}. This record does not assert independent reproduction.

### GDiff

DF ROUGE: 41.9. Result ID: 6a1d43c1fbc373592c167791.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11e2dbd7150506f83de5.md>)

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

Recorded evaluation configuration: {"Backbone":"Llama2-7B"}. This record does not assert independent reproduction.

### DareU

DF ROUGE: 40.7. Result ID: 6a1d43c1fbc373592c167796.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11e2dbd7150506f83de5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a1d11e2dbd7150506f83de5.md>)

Recorded evaluation configuration: {"Backbone":"Llama2-7B"}. This record does not assert independent reproduction.

### Retraining

DF ROUGE: 38.1. Result ID: 6a1d43c1fbc373592c16778e.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11e2dbd7150506f83de5.md>)

Recorded evaluation configuration: {"Backbone":"Llama2-7B"}. This record does not assert independent reproduction.

### SCRN

DF ROUGE: 29.2. Result ID: 6a1d43c1fbc373592c167793.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11e2dbd7150506f83de5.md>)

Recorded evaluation configuration: {"Backbone":"Llama2-7B"}. This record does not assert independent reproduction.

### NPO

DF ROUGE: 28.6. Result ID: 6a1d43c1fbc373592c167795.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11e2dbd7150506f83de5.md>)

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

Recorded evaluation configuration: {"Backbone":"Llama2-7B"}. This record does not assert independent reproduction.

### SCRUB

DF ROUGE: 26.8. Result ID: 6a1d43c1fbc373592c167792.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11e2dbd7150506f83de5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71011e209131cea4e7b9.md>)

[Implementation](<https://github.com/Meghdad92/SCRUB>)

Recorded evaluation configuration: {"Backbone":"Llama2-7B"}. This record does not assert independent reproduction.

### IDK

DF ROUGE: 2.8. Result ID: 6a1d43c1fbc373592c167794.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11e2dbd7150506f83de5.md>)

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

Recorded evaluation configuration: {"Backbone":"Llama2-7B"}. This record does not assert independent reproduction.

### GA

DF ROUGE: 0. Result ID: 6a1d43c1fbc373592c167790.

[Reporting paper](<https://www.sota2.com/research/paper/6a1d11e2dbd7150506f83de5.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa71101e209131cea4efdf.md>)

Recorded evaluation configuration: {"Backbone":"Llama2-7B"}. This record does not assert independent reproduction.

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