HelpSteer
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
HelpSteer2 (test)
0.8Avg Pref Score vs QWEN2.5-0.5B
5
Helpsteer 13B+ Models Holdout (test)
74.1Acc_pair (RM1 Helpful)
4
Helpsteer 30B+ Models Holdout (test)
76.5Pairwise Accuracy (RM1)
4
Helpsteer 70B+ Models Holdout (test)
77.8Pairwise Acc (RM1)
4
Helpsteer 10% holdout (test)
85.5Pairwise Acc (RM1-Helpful)
4
Helpsteer 5% holdout (test)
84.9Pairwise Accuracy (RM1-Helpful)
4
Helpsteer 2% holdout (test)
86.6Pairwise Acc (RM1)
4
HelpSteer2
3FA Rate
3
HelpSteer2
12.66Hypervolume (HV)
3
HelpSteer2
16.81HV
3
HelpSteer1 sampled (train)
3.89Usefulness Score
2