List your company

Preference Classification on Helpfulness & Harmlessness (HH) (test)

70.4Classification AccuracyBTPO
62.39264.47166.5568.629Oct 17, 2025

Evaluation results

MethodLinks
BTPOBackbone=Qwen2.5-7B-Instruct2025.1070.4
GRPO (pair)Backbone=Qwen2.5-7B-Instruct2025.1064.7
RM-R1Backbone=Qwen2.5-7B-Instruct2025.1062.7
RM-R1-finetunedBackbone=Qwen2.5-7B-Instruct, Fine-tuned=true2025.1062.7