# Machine Unlearning on MUSE forget-set (Quantization Metrics)

Source: [Machine Unlearning on MUSE forget-set (Quantization Metrics)](<https://www.sota2.com/research/sota/machine-unlearning-on-muse-forget-set-quantization-metrics>)

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

Data updated: 2026-05-15T09:59:37.446Z

Evaluation of Machine Unlearning performance on the MUSE forget-set, focusing on metrics related to model quantization techniques (BF16, NF4) and the resulting performance delta (∆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) | MUSE forget-set |
| Metric | Performance (BF16) |
| Unit | SCALAR |
| 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

- [Performance (BF16)](<https://www.sota2.com/research/sota/machine-unlearning-on-muse-forget-set-quantization-metrics.md?metric=Performance%20(BF16)>): Higher is better; unit: SCALAR.

- [Performance (NF4)](<https://www.sota2.com/research/sota/machine-unlearning-on-muse-forget-set-quantization-metrics.md?metric=Performance%20(NF4)>): Higher is better; unit: SCALAR.

- [Performance Delta (PTQ)](<https://www.sota2.com/research/sota/machine-unlearning-on-muse-forget-set-quantization-metrics.md?metric=Performance%20Delta%20(PTQ)>): Lower is better; unit: SCALAR.

## Recorded results

| Method | Performance (BF16) | Result date | Report paper |
| --- | --- | --- | --- |
| Zero-shot | 0.365 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| LUNAR | 0.187 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| LUNAR | 0.162 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| Zero-shot | 0.024 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| MANSU | 0.021 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| Surgical GA | 0.02 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| NPO | 0.013 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| GU+SimNPO | 0.013 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| NPO | 0.011 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| MANSU | 0.005 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| SimNPO | 0.001 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| Global GA | 0 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| Surgical GA | 0 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| SimNPO | 0 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| GU+SimNPO | 0 | Not recorded | 6a06a8c9dbc836b87d82a365 |
| Global GA | 0 | Not recorded | 6a06a8c9dbc836b87d82a365 |

### Zero-shot

Performance (BF16): 0.365. Result ID: 6a06ee7247f34ce924ef1c6f.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction.

### LUNAR

Performance (BF16): 0.187. Result ID: 6a06ee7247f34ce924ef1c75.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bf31e209131cea8b861.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction.

### LUNAR

Performance (BF16): 0.162. Result ID: 6a06ee7247f34ce924ef1c7d.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bf31e209131cea8b861.md>)

Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction.

### Zero-shot

Performance (BF16): 0.024. Result ID: 6a06ee7247f34ce924ef1c77.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction.

### MANSU

Performance (BF16): 0.021. Result ID: 6a06ee7247f34ce924ef1c7e.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction.

### Surgical GA

Performance (BF16): 0.02. Result ID: 6a06ee7247f34ce924ef1c79.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e421e209131cea3ec45.md>)

[Implementation](<https://github.com/joeljang/knowledge-unlearning>)

Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction.

### NPO

Performance (BF16): 0.013. Result ID: 6a06ee7247f34ce924ef1c72.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

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

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction.

### GU+SimNPO

Performance (BF16): 0.013. Result ID: 6a06ee7247f34ce924ef1c7c.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75011e209131cea652f5.md>)

Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction.

### NPO

Performance (BF16): 0.011. Result ID: 6a06ee7247f34ce924ef1c7a.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

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

Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction.

### MANSU

Performance (BF16): 0.005. Result ID: 6a06ee7247f34ce924ef1c76.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction.

### SimNPO

Performance (BF16): 0.001. Result ID: 6a06ee7247f34ce924ef1c7b.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c3d1e209131cea8d186.md>)

Recorded evaluation configuration: {"Backbone":"Qwen-3-8B"}. This record does not assert independent reproduction.

### Global GA

Performance (BF16): 0. Result ID: 6a06ee7247f34ce924ef1c70.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e421e209131cea3ec45.md>)

[Implementation](<https://github.com/joeljang/knowledge-unlearning>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction.

### Surgical GA

Performance (BF16): 0. Result ID: 6a06ee7247f34ce924ef1c71.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e421e209131cea3ec45.md>)

[Implementation](<https://github.com/joeljang/knowledge-unlearning>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction.

### SimNPO

Performance (BF16): 0. Result ID: 6a06ee7247f34ce924ef1c73.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c3d1e209131cea8d186.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction.

### GU+SimNPO

Performance (BF16): 0. Result ID: 6a06ee7247f34ce924ef1c74.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75011e209131cea652f5.md>)

Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction.

### Global GA

Performance (BF16): 0. Result ID: 6a06ee7247f34ce924ef1c78.

[Reporting paper](<https://www.sota2.com/research/paper/6a06a8c9dbc836b87d82a365.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6e421e209131cea3ec45.md>)

[Implementation](<https://github.com/joeljang/knowledge-unlearning>)

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.
