# Non-toxic generation on RealToxicityPrompts

Source: [Non-toxic generation on RealToxicityPrompts](<https://www.sota2.com/research/sota/non-toxic-generation-on-realtoxicityprompts>)

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

Data updated: 2026-05-19T07:11:10.031Z

Evaluation of model capability to generate text that avoids toxicity using the RealToxicityPrompts dataset, measured by average maximum toxicity, toxicity probability, and generation perplexity.

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 | Non-toxic generation |
| Dataset (including any stated split) | RealToxicityPrompts |
| Metric | Avg. Max Toxicity |
| Unit | SCALAR |
| Direction | Lower is better |
| Results loaded | 8 of 8 |
| 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

- [Avg. Max Toxicity](<https://www.sota2.com/research/sota/non-toxic-generation-on-realtoxicityprompts.md?metric=Avg.%20Max%20Toxicity>): Lower is better; unit: SCALAR.

- [Toxicity Probability](<https://www.sota2.com/research/sota/non-toxic-generation-on-realtoxicityprompts.md?metric=Toxicity%20Probability>): Lower is better; unit: SCALAR.

- [Perplexity](<https://www.sota2.com/research/sota/non-toxic-generation-on-realtoxicityprompts.md?metric=Perplexity>): Lower is better; unit: SCALAR.

## Recorded results

| Method | Avg. Max Toxicity | Result date | Report paper |
| --- | --- | --- | --- |
| GCSL-bey-NL | 0.115 | Not recorded | 6a0bee83dbc836b87d82c32b |
| GCSL-bey | 0.125 | Not recorded | 6a0bee83dbc836b87d82c32b |
| GCSL-NL | 0.129 | Not recorded | 6a0bee83dbc836b87d82c32b |
| LLMEraser | 0.13 | Not recorded | 6a0bee83dbc836b87d82c32b |
| GCSL | 0.134 | Not recorded | 6a0bee83dbc836b87d82c32b |
| SFT | 0.139 | Not recorded | 6a0bee83dbc836b87d82c32b |
| DEXPERT | 0.145 | Not recorded | 6a0bee83dbc836b87d82c32b |
| PPLM | 0.152 | Not recorded | 6a0bee83dbc836b87d82c32b |

### GCSL-bey-NL

Avg. Max Toxicity: 0.115. Result ID: 6a0c0be6fbc373592c019453.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee83dbc836b87d82c32b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee83dbc836b87d82c32b.md>)

Recorded evaluation configuration: {"goal definition":"beyond-threshold","goal representation":"natural language"}. This record does not assert independent reproduction.

### GCSL-bey

Avg. Max Toxicity: 0.125. Result ID: 6a0c0be6fbc373592c019452.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee83dbc836b87d82c32b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee83dbc836b87d82c32b.md>)

Recorded evaluation configuration: {"goal definition":"beyond-threshold"}. This record does not assert independent reproduction.

### GCSL-NL

Avg. Max Toxicity: 0.129. Result ID: 6a0c0be6fbc373592c019451.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee83dbc836b87d82c32b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee83dbc836b87d82c32b.md>)

Recorded evaluation configuration: {"goal representation":"natural language"}. This record does not assert independent reproduction.

### LLMEraser

Avg. Max Toxicity: 0.13. Result ID: 6a0c0be6fbc373592c01944f.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee83dbc836b87d82c32b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa781b1e209131cea76d42.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### GCSL

Avg. Max Toxicity: 0.134. Result ID: 6a0c0be6fbc373592c019450.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee83dbc836b87d82c32b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a0bee83dbc836b87d82c32b.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### SFT

Avg. Max Toxicity: 0.139. Result ID: 6a0c0be6fbc373592c01944c.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee83dbc836b87d82c32b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa76e01e209131cea6f85d.md>)

Recorded evaluation configuration: {"Training samples":"top 25% non-toxic"}. This record does not assert independent reproduction.

### DEXPERT

Avg. Max Toxicity: 0.145. Result ID: 6a0c0be6fbc373592c01944d.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee83dbc836b87d82c32b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6a0b1e209131cea25b2e.md>)

[Implementation](<https://github.com/alisawuffles/DExperts>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### PPLM

Avg. Max Toxicity: 0.152. Result ID: 6a0c0be6fbc373592c01944e.

[Reporting paper](<https://www.sota2.com/research/paper/6a0bee83dbc836b87d82c32b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67241e209131cea149df.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

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