# Activation Steering on AxBench Gemma-2-9B layer 20

Source: [Activation Steering on AxBench Gemma-2-9B layer 20](<https://www.sota2.com/research/sota/activation-steering-on-axbench-gemma-2-9b-layer-20>)

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

Data updated: 2026-05-06T08:57:36.192Z

Evaluation of Activation Steering performance on layer 20 of the Gemma-2-9B model using the AxBench dataset, measured by the Steering score.

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 | Activation Steering |
| Dataset (including any stated split) | AxBench Gemma-2-9B layer 20 |
| Metric | Steering Score |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 17 of 17 |
| 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

- [Steering Score](<https://www.sota2.com/research/sota/activation-steering-on-axbench-gemma-2-9b-layer-20.md?metric=Steering%20Score>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Steering Score | Result date | Report paper |
| --- | --- | --- | --- |
| A-PSRMSE | 1.12 | Not recorded | 69facb48c0f56ed84308a5ef |
| Hyper-Steer | 1.091 | Not recorded | 69facb48c0f56ed84308a5ef |
| prompt | 1.075 | Not recorded | 69facb48c0f56ed84308a5ef |
| A-ConstMSE | 1.053 | Not recorded | 69facb48c0f56ed84308a5ef |
| S-ConstMSE | 0.903 | Not recorded | 69facb48c0f56ed84308a5ef |
| S-PSRMSE | 0.9 | Not recorded | 69facb48c0f56ed84308a5ef |
| ΦSV,r=1 | 0.892 | Not recorded | 69facb48c0f56ed84308a5ef |
| A-PSRLL | 0.827 | Not recorded | 69facb48c0f56ed84308a5ef |
| LoReFT-RePS | 0.757 | Not recorded | 69facb48c0f56ed84308a5ef |
| A-ConstLL | 0.757 | Not recorded | 69facb48c0f56ed84308a5ef |
| S-PSRLL | 0.667 | Not recorded | 69facb48c0f56ed84308a5ef |
| S-ConstLL | 0.633 | Not recorded | 69facb48c0f56ed84308a5ef |
| LoRA-RePS | 0.631 | Not recorded | 69facb48c0f56ed84308a5ef |
| ReFT-r1 | 0.63 | Not recorded | 69facb48c0f56ed84308a5ef |
| LoRA | 0.602 | Not recorded | 69facb48c0f56ed84308a5ef |
| S-ConstDiM | 0.322 | Not recorded | 69facb48c0f56ed84308a5ef |
| SAE | 0.191 | Not recorded | 69facb48c0f56ed84308a5ef |

### A-PSRMSE

Steering Score: 1.12. Result ID: 69fb027c47f34ce924de1326.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

Recorded evaluation configuration: {"Method Type":"Multi-rank and/or multi-layer"}. This record does not assert independent reproduction.

### Hyper-Steer

Steering Score: 1.091. Result ID: 69fb027c47f34ce924de1322.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa79111e209131cea7bf3d.md>)

Recorded evaluation configuration: {"Method Type":"Multi-rank and/or multi-layer"}. This record does not assert independent reproduction.

### prompt

Steering Score: 1.075. Result ID: 69fb027c47f34ce924de131e.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d18.md>)

Recorded evaluation configuration: {"Method Type":"Multi-rank and/or multi-layer"}. This record does not assert independent reproduction.

### A-ConstMSE

Steering Score: 1.053. Result ID: 69fb027c47f34ce924de1325.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

Recorded evaluation configuration: {"Method Type":"Multi-rank and/or multi-layer"}. This record does not assert independent reproduction.

### S-ConstMSE

Steering Score: 0.903. Result ID: 69fb027c47f34ce924de1312.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

Recorded evaluation configuration: {"Rank":"1","Intervention type":"single-layer"}. This record does not assert independent reproduction.

### S-PSRMSE

Steering Score: 0.9. Result ID: 69fb027c47f34ce924de1313.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

Recorded evaluation configuration: {"Rank":"1","Intervention type":"single-layer"}. This record does not assert independent reproduction.

### ΦSV,r=1

Steering Score: 0.892. Result ID: 69fb027c47f34ce924de130f.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78e11e209131cea7af72.md>)

Recorded evaluation configuration: {"Rank":"1","Intervention type":"single-layer"}. This record does not assert independent reproduction.

### A-PSRLL

Steering Score: 0.827. Result ID: 69fb027c47f34ce924de1324.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

Recorded evaluation configuration: {"Method Type":"Multi-rank and/or multi-layer"}. This record does not assert independent reproduction.

### LoReFT-RePS

Steering Score: 0.757. Result ID: 69fb027c47f34ce924de1321.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78e11e209131cea7af72.md>)

Recorded evaluation configuration: {"Method Type":"Multi-rank and/or multi-layer"}. This record does not assert independent reproduction.

### A-ConstLL

Steering Score: 0.757. Result ID: 69fb027c47f34ce924de1323.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

Recorded evaluation configuration: {"Method Type":"Multi-rank and/or multi-layer"}. This record does not assert independent reproduction.

### S-PSRLL

Steering Score: 0.667. Result ID: 69fb027c47f34ce924de1311.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

Recorded evaluation configuration: {"Rank":"1","Intervention type":"single-layer"}. This record does not assert independent reproduction.

### S-ConstLL

Steering Score: 0.633. Result ID: 69fb027c47f34ce924de1310.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

Recorded evaluation configuration: {"Rank":"1","Intervention type":"single-layer"}. This record does not assert independent reproduction.

### LoRA-RePS

Steering Score: 0.631. Result ID: 69fb027c47f34ce924de1320.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78e11e209131cea7af72.md>)

Recorded evaluation configuration: {"Method Type":"Multi-rank and/or multi-layer"}. This record does not assert independent reproduction.

### ReFT-r1

Steering Score: 0.63. Result ID: 69fb027c47f34ce924de130e.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d18.md>)

Recorded evaluation configuration: {"Rank":"1","Intervention type":"single-layer"}. This record does not assert independent reproduction.

### LoRA

Steering Score: 0.602. Result ID: 69fb027c47f34ce924de131f.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d18.md>)

Recorded evaluation configuration: {"Method Type":"Multi-rank and/or multi-layer"}. This record does not assert independent reproduction.

### S-ConstDiM

Steering Score: 0.322. Result ID: 69fb027c47f34ce924de130c.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d18.md>)

Recorded evaluation configuration: {"Rank":"1","Intervention type":"single-layer"}. This record does not assert independent reproduction.

### SAE

Steering Score: 0.191. Result ID: 69fb027c47f34ce924de130d.

[Reporting paper](<https://www.sota2.com/research/paper/69facb48c0f56ed84308a5ef.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa773a1e209131cea71d18.md>)

Recorded evaluation configuration: {"Rank":"1","Intervention type":"single-layer"}. This record does not assert independent reproduction.

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