List your company

Mathematical Problem Solving on OlympiadBench (675 samples)

19.37AccuracyHAPO
13.587615.088816.5918.0912Sep 20, 2025

Evaluation results

MethodLinks
HAPOBackbone=LLaMA3.2-3B Instruct2025.0919.37
HAPOBackbone=LLaMA3.1-8B-Instruct2025.0917.66
DAPO w/ Forking TokensBackbone=LLaMA3.2-3B Instruct2025.0916.89
Vanilla DAPOBackbone=LLaMA3.2-3B Instruct2025.0916.63
Vanilla DAPOBackbone=LLaMA3.1-8B-Instruct2025.0916.03
DAPO w/ Forking TokensBackbone=LLaMA3.1-8B-Instruct2025.0915.89
Vanilla GRPOBackbone=LLaMA3.2-3B Instruct2025.0914.35
Vanilla GRPOBackbone=LLaMA3.1-8B-Instruct2025.0913.81