HelpSteer
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
HelpSteer (test)
0.077MAE
65
HelpSteer 3
83.15Accuracy
62
Helpsteer
0.72L1 Error
48
HelpSteer2 (test)
18.7WCE
48
HelpSteer2
0.987Diversity
36
HelpSteer2
0.464Spearman Correlation
28
HelpSteer (test)
8.34AlpacaEval 2 WR
27
HelpSteer
3.89Helpfulness
22
HelpSteer2
99.86Harmless Rate
20
HelpSteer
0.51Creative ArmoRM Score
18
HelpSteer3 (H3) v1 (test)
8.96AOG
16
HelpSteer2
72.3Accuracy
16
HelpSteer
99.5Creative Win Rate
15
HelpSteer
0.543R2@50
14
HelpSteer2
0.507Helpfulness
14
HelpSteer 1
51.83Helpfulness
14
HelpSteer2 Relative Positive Representative Target
0.946Diversity
12
HelpSteer3
-5.89Score
10
HelpSteer2
0.65Spearman Correlation
10
HelpSteer2 (In-Distribution)
0.342Kendall's Tau
9
HelpSteer2 held-out (test)
68.4Preference Accuracy
7
HelpSteer-2
0.03CV (Helpfulness)
7
HelpSteer2 (test)
0.02GPU Hours
6
HelpSteer (test)
61.25Pairwise Win Rate (excl. ties)
5
HelpSteer3 (test)
-5.89Score
5