List your company
ResearchFormerly WizWand

Step-level correctness assessment on MATH (test)

58.6PR-AUCQwen2.5-Math-7B-PRM800k
14.71226.10637.548.894Nov 9, 2025
Updated 6mo ago

Evaluation results

MethodLinks
Qwen2.5-Math-7B-PRM800k# Sample=265K2025.1158.6
ReProbe# Sample=32K, Architecture=Hidden States, Annotation=Self-anno2025.1155.8
Universal-PRM-Qwen2.5-Math-7B# Sample=690K2025.1153.4
ReProbe# Sample=32K, Architecture=Hidden States, Annotation=DeepSeek-anno2025.1153.4
Qwen2.5-Math-PRM-7B# Sample=860K2025.1153.1
ReProbe# Sample=32K, Architecture=Attn+Logit, Annotation=Self-anno2025.1152.9
ReProbe# Sample=32K, Architecture=Attn+Logit, Annotation=DeepSeek-anno2025.1146.5
Math-Shepherd-PRM-7B# Sample=440K2025.1138
RLHFlow-PRM-Deepseek-8B# Sample=253K2025.1128.9
Skywork-PRM-1.5B# Sample=Unk2025.1128.3
H4-Qwen2.5-PRM-1.5B-0.2# Sample=369K2025.1125.9
Semantic Entropy# Sample=-2025.1125.7
CCP# Sample=-2025.1125
Lexical Similarity# Sample=-2025.1125
RLHFlow-PRM-Mistral-8B# Sample=273K2025.1123.3
Degree Matrix# Sample=-2025.1122.7
MaxProb# Sample=-2025.1122.1
Self-Certainty# Sample=-2025.1121.3
MaxEntropy# Sample=-2025.1121.2
Perplexity# Sample=-2025.1120.5
Random# Sample=-2025.1117.3
P(True)# Sample=-2025.1116.4