RealToxicityPrompts
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
RealToxicityPrompts
103.6Relative Throughput
126
RealToxicityPrompts (test)
91.3Distinct-1
54
RealToxicityPrompts challenging
6.2Avg Toxicity (Max)
46
RealToxicityPrompts
0Toxicity Score
33
RealToxicityPrompts challenging
0.062Max Toxicity
32
REALTOXICITYPROMPTS
21.24Toxicity
24
RealToxicityPrompts
0.27Avg Max Toxicity
22
RealToxicityPrompts 1k samples
0.51CLS Toxicity
20
RealToxicityPrompts (test)
90.11AUC
20
RealToxicityPrompts 1K non-toxic prompts, 1K toxic prompts
5Count of Non-Toxic Samples
14
RealToxicityPrompts (test)
10.1Full Toxicity
14
RealToxicityPrompts (RTP) (test)
31.36Overall Score
12
RealToxicityPrompts (RTP)
0.53CLS Tox Rate
12
RealToxicityPrompts (test)
9.2Perspective API Score
12
RealToxicityPrompts Nontoxic
0.22Exp. Max. Toxicity
10
RealToxicityPrompts 10K nontoxic prompts
30.2Avg Max Toxicity
9
RealToxicityPrompts
0.115Avg. Max Toxicity
8
RealToxicityPrompts malicious
14.8Attack Success Rate (ASR)
8
RealToxicityPrompts (hold-out)
4.97Detoxify Identity Attack Score
7
RealToxicityPrompts primary evaluation
3.05Detoxify Identity Attack Score
7
RealToxicityPrompts 10K nontoxic prompts GPT2-large generation (test)
0.172Max Toxicity
7
RealToxicityPrompts sampled completions
0.475Spearman Correlation Coefficient
6
RealToxicityPrompts
90Coverage
5
RealToxicityPrompts
23Identity Count
5
RealToxicityPrompts RTP-N (Nontoxic)
0.2Toxic Fraction
5