List your company
ResearchFormerly WizWand

Mathematical Reasoning on Minerva-Math (Last, Best)

49.63Last ScoreHT-MNPO
-1.88192811.49136124.8646538.237939May 23, 2025Jun 13, 2025Jul 4, 2025Jul 25, 2025Aug 15, 2025Sep 5, 2025Sep 27, 2025
Updated 7mo ago

Evaluation results

MethodLinks
HT-MNPOReward Model=Skywork-Reward-V22025.0949.63—
HT-MNPOReward Model=ArmoRM-Llama32025.0947.79—
SPPO2025.0947.43—
HT-MNPOReward Model=Athene-RM-8B2025.0947.06—
INPO2025.0946.32—
DPO2025.0945.96—
TD-MNPO2025.0944.85—
SFT Model2025.0944.12—
SimPO2025.0943.38—
REINFORCE++Context Length=4k, Base Model=Qwen-2.5-7B-Instruct2025.050.16910.1728
RPG-URKLContext Length=4k, Base Model=Qwen-2.5-7B-Instruct2025.050.16540.1654
RPG-UFKLContext Length=4k, Base Model=Qwen-2.5-7B-Instruct2025.050.13970.2059
RPG-REINFORCE-UFKLContext Length=4k, Base Model=Qwen-2.5-7B-Instruct2025.050.1360.1434
REINFORCE++-BaselineContext Length=4k, Base Model=Qwen-2.5-7B-Instruct2025.050.1250.1471
RPG-REINFORCE-URKLContext Length=4k, Base Model=Qwen-2.5-7B-Instruct2025.050.1140.1434
GRPOContext Length=4k, Base Model=Qwen-2.5-7B-Instruct2025.050.10290.1177
DAPOContext Length=4k, Base Model=Qwen-2.5-7B-Instruct2025.050.09930.1507