# Toxicity mitigation on ToxicRandom (test)

Source: [Toxicity mitigation on ToxicRandom (test)](<https://www.sota2.com/research/sota/toxicity-mitigation-on-toxicrandom-test>)

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

Data updated: 2026-02-26T05:57:54.244Z

Evaluation of model performance on the test split of the ToxicRandom dataset, focusing on toxicity mitigation capabilities, measured by Perplexity and Toxicity 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 | Toxicity mitigation |
| Dataset (including any stated split) | ToxicRandom (test) |
| Metric | Perplexity |
| Unit | SCALAR |
| Direction | Lower is better |
| Results loaded | 6 of 6 |
| 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

- [Perplexity](<https://www.sota2.com/research/sota/toxicity-mitigation-on-toxicrandom-test.md?metric=Perplexity>): Lower is better; unit: SCALAR.

- [Toxicity](<https://www.sota2.com/research/sota/toxicity-mitigation-on-toxicrandom-test.md?metric=Toxicity>): Lower is better; unit: PERCENT.

## Recorded results

| Method | Perplexity | Result date | Report paper |
| --- | --- | --- | --- |
| CONTINUATION | 27.21 | Not recorded | 68fa734d1e209131cea5ba89 |
| INJECTION | 30.57 | Not recorded | 68fa734d1e209131cea5ba89 |
| DATG-L | 31.54 | Not recorded | 68fa734d1e209131cea5ba89 |
| DATG-P | 41.28 | Not recorded | 68fa734d1e209131cea5ba89 |
| PREADD | 47.39 | Not recorded | 68fa734d1e209131cea5ba89 |
| FUDGE | 53.29 | Not recorded | 68fa734d1e209131cea5ba89 |

### CONTINUATION

Perplexity: 27.21. Result ID: 6992b7063a6c31484910f45e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa734d1e209131cea5ba89.md>)

Recorded evaluation configuration: {"Generator":"CONTINUATION"}. This record does not assert independent reproduction.

### INJECTION

Perplexity: 30.57. Result ID: 6992b7063a6c31484910f45f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa734d1e209131cea5ba89.md>)

Recorded evaluation configuration: {"Generator":"INJECTION"}. This record does not assert independent reproduction.

### DATG-L

Perplexity: 31.54. Result ID: 6992b7063a6c31484910f462.

[Reporting paper](<https://www.sota2.com/research/paper/68fa734d1e209131cea5ba89.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa734d1e209131cea5ba89.md>)

[Implementation](<https://github.com/IAAR-Shanghai/DATG>)

Recorded evaluation configuration: {"Generator":"DATG-L"}. This record does not assert independent reproduction.

### DATG-P

Perplexity: 41.28. Result ID: 6992b7063a6c31484910f463.

[Reporting paper](<https://www.sota2.com/research/paper/68fa734d1e209131cea5ba89.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa734d1e209131cea5ba89.md>)

[Implementation](<https://github.com/IAAR-Shanghai/DATG>)

Recorded evaluation configuration: {"Generator":"DATG-P"}. This record does not assert independent reproduction.

### PREADD

Perplexity: 47.39. Result ID: 6992b7063a6c31484910f461.

[Reporting paper](<https://www.sota2.com/research/paper/68fa734d1e209131cea5ba89.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6ff81e209131cea483ef.md>)

[Implementation](<https://github.com/jonnypei/acl23-preadd>)

Recorded evaluation configuration: {"Generator":"PREADD"}. This record does not assert independent reproduction.

### FUDGE

Perplexity: 53.29. Result ID: 6992b7063a6c31484910f460.

[Reporting paper](<https://www.sota2.com/research/paper/68fa734d1e209131cea5ba89.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a931e209131cea29040.md>)

Recorded evaluation configuration: {"Generator":"FUDGE"}. This record does not assert independent reproduction.

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