List your company
ResearchFormerly WizWand

Mathematical Reasoning on Minerva (Mean@8)

38Mean@8GSPO
7.517615.431323.34531.2587May 17, 2026May 23, 2026May 29, 2026Jun 4, 2026Jun 10, 2026Jun 16, 2026Jun 22, 2026
Updated 4mo ago

Evaluation results

MethodLinks
GSPOBackbone=Qwen3-4B-Base2026.0538
DISABackbone=Qwen3-4B-Base2026.0537.4
FlowRLBackbone=Qwen3-4B-Base2026.0536.8
FlowRLBackbone=Qwen2.5-7B2026.0534.1
Offline KDStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-4B2026.0633.23
AsymPOModel=Qwen3-1.7B-Base, Training Data=4k subset of DAPO-17K, Evaluation Rollouts=82026.0630.65
SPOModel=Qwen3-1.7B-Base, Training Data=4k subset of DAPO-17K, Evaluation Rollouts=82026.0630
GRPOBackbone=Qwen3-4B-Base2026.0529
DEARStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-4B2026.0628.68
Standard OPDStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-4B2026.0627.94
DISABackbone=Qwen2.5-7B2026.0527.9
GRPOModel=Qwen3-1.7B-Base, Training Data=4k subset of DAPO-17K, Evaluation Rollouts=82026.0627.53
VanillaBackbone=Qwen3-4B-Base2026.0526.7
GSPOBackbone=Qwen2.5-7B2026.0522.7
GRPOBackbone=Qwen2.5-7B2026.0522.4
VanillaBackbone=Qwen2.5-7B2026.0522.2
Base modelStudent Model=Qwen3-1.7B, Teacher Model=Qwen3-4B2026.0620.31
Offline KDStudent Model=Qwen2.5-1.5B, Teacher Model=Qwen2.5-14B2026.0614.57
DEARStudent Model=Qwen2.5-1.5B, Teacher Model=Qwen2.5-14B2026.0613.83
DEARStudent Model=Qwen2.5-1.5B, Teacher Model=Qwen3-4B2026.0613.6
Offline KDStudent Model=Qwen2.5-1.5B, Teacher Model=Qwen3-4B2026.0612.82
Standard OPDStudent Model=Qwen2.5-1.5B, Teacher Model=Qwen2.5-14B2026.0612.73
Standard OPDStudent Model=Qwen2.5-1.5B, Teacher Model=Qwen3-4B2026.0611.63
Base modelStudent Model=Qwen2.5-1.5B, Teacher Model=Qwen2.5-14B2026.069.1
Base modelStudent Model=Qwen2.5-1.5B, Teacher Model=Qwen3-4B2026.068.69