List your company

Math Reasoning on GSM DE

66AccuracyKL-DRTO
46.86451.83256.861.768Mar 27, 2026

Evaluation results

MethodLinks
KL-DRTOBackbone=Llama3-8B2026.0366
χ2-DRTOBackbone=Llama3-8B2026.0365.6
PPOBackbone=Llama3-8B2026.0362
DPOBackbone=Llama3-8B2026.0362
RTOBackbone=Llama3-8B2026.0360.8
GRPOBackbone=Llama3-8B2026.0360
SFTBackbone=Llama3-8B2026.0347.6