List your company

Short-form Question Answering on PopQA (test)

36.34Exact MatchVSPO + PRS_short
7.37614.895522.41529.9345Dec 8, 2025

Evaluation results

MethodLinks
VSPO + PRS_shortBackbone=Qwen2.5-3B-Instruct2025.1236.34
GRPO + PRS_shortBackbone=Qwen2.5-3B-Instruct2025.1235.62
PPO + PRS_shortBackbone=Qwen2.5-3B-Instruct2025.1235.62
SFTBackbone=Qwen2.5-3B-Instruct2025.1210.4
UntrainedBackbone=Qwen2.5-3B-Instruct2025.128.49