List your company
ResearchFormerly WizWand

Mathematical Reasoning on Minerva-Math (Acc, Len)

38.97AccuracyLCPO
17.941223.400628.8634.3194Aug 5, 2025Aug 6, 2025Aug 7, 2025Aug 9, 2025Aug 10, 2025Aug 11, 2025Aug 13, 2025
Updated 4mo ago

Evaluation results

MethodLinks
LCPOBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0838.972,079
TrEffBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0837.132,917
OriginalBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0836.765,926
CoDBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0836.42,733
L1-MaxBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0836.41,908
L1-ExactBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0836.033,341
DASTBackbone=DeepSeek-R1-Distill-Qwen-7B2025.0836.038,119
L1-MaxBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.0827.573,418
LCPOBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.0827.212,596
OriginalBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.0826.847,211
CoDBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.0825.746,231
TrEffBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.0825.742,950
L1-ExactBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.0825.373,673
VRPOOptimization Type=test-time optimization, Reward Type=rule-based rewards2025.0820.22—
GRPOOptimization Type=test-time optimization, Reward Type=rule-based rewards2025.0819.49—
PPOOptimization Type=test-time optimization, Reward Type=rule-based rewards2025.0819.49—
Cold StartOptimization Type=test-time optimization, Reward Type=rule-based rewards2025.0818.75—