List your company

Step-level correctness assessment on GSM8k (test)

70.2PR-AUCQwen2.5-Math-PRM-7B
3.32820.68938.0555.411Nov 9, 2025

Evaluation results

MethodLinks
Qwen2.5-Math-PRM-7B# Sample=860K2025.1170.2
ReProbe# Sample=32K, Architecture=Hidden States, Annotation=Self-anno2025.1167.3
ReProbe# Sample=32K, Architecture=Hidden States, Annotation=DeepSeek-anno2025.1165
Universal-PRM-Qwen2.5-Math-7B# Sample=690K2025.1162.4
ReProbe# Sample=32K, Architecture=Attn+Logit, Annotation=DeepSeek-anno2025.1161.6
Qwen2.5-Math-7B-PRM800k# Sample=265K2025.1161.3
ReProbe# Sample=32K, Architecture=Attn+Logit, Annotation=Self-anno2025.1159.4
RLHFlow-PRM-Deepseek-8B# Sample=253K2025.1154
RLHFlow-PRM-Mistral-8B# Sample=273K2025.1153.7
Skywork-PRM-1.5B# Sample=Unk2025.1141.2
Math-Shepherd-PRM-7B# Sample=440K2025.1140.5
H4-Qwen2.5-PRM-1.5B-0.2# Sample=369K2025.1117.1
MaxEntropy# Sample=-2025.1112.2
Lexical Similarity# Sample=-2025.1111.9
Semantic Entropy# Sample=-2025.1111.6
MaxProb# Sample=-2025.1110.6
Self-Certainty# Sample=-2025.1110.1
Perplexity# Sample=-2025.119.9
CCP# Sample=-2025.119
Degree Matrix# Sample=-2025.118.9
Random# Sample=-2025.116.1
P(True)# Sample=-2025.115.9