List your company

Step-level correctness assessment on Meetings (test)

81.9PR-AUCReProbe
52.05259.80167.5575.299Nov 9, 2025

Evaluation results

MethodLinks
ReProbe# Sample=32K, Architecture=Hidden States, Annotation=Self-anno2025.1181.9
ReProbe# Sample=32K, Architecture=Hidden States, Annotation=DeepSeek-anno2025.1181.7
ReProbe# Sample=32K, Architecture=Attn+Logit, Annotation=DeepSeek-anno2025.1180.2
ReProbe# Sample=32K, Architecture=Attn+Logit, Annotation=Self-anno2025.1177.9
Qwen2.5-Math-7B-PRM800k# Sample=265K2025.1176.8
Qwen2.5-Math-PRM-7B# Sample=860K2025.1175.7
Universal-PRM-Qwen2.5-Math-7B# Sample=690K2025.1175.3
Math-Shepherd-PRM-7B# Sample=440K2025.1166
MaxProb# Sample=-2025.1165.5
CCP# Sample=-2025.1164.5
Self-Certainty# Sample=-2025.1164.3
H4-Qwen2.5-PRM-1.5B-0.2# Sample=369K2025.1163.3
MaxEntropy# Sample=-2025.1161.8
Semantic Entropy# Sample=-2025.1161
P(True)# Sample=-2025.1160.8
Lexical Similarity# Sample=-2025.1160.3
Degree Matrix# Sample=-2025.1159.7
Random# Sample=-2025.1158.8
RLHFlow-PRM-Deepseek-8B# Sample=253K2025.1157.9
Perplexity# Sample=-2025.1157.2
RLHFlow-PRM-Mistral-8B# Sample=273K2025.1155.5
Skywork-PRM-1.5B# Sample=Unk2025.1153.2