List your company

Math Reasoning on LIMR

96.59Relative Overall ScoreReverse KL
-3.863622.215748.29574.3743Feb 5, 2026

Evaluation results

MethodLinks
Reverse KLBackbone=Qwen2.5-Math-7B2026.0296.592.4
PearsonBackbone=Qwen2.5-Math-7B2026.0296.282.8
Jensen-ShannonBackbone=Qwen2.5-Math-7B2026.0294.933.8
KLBackbone=Qwen2.5-Math-7B2026.0293.643.4
GRPOBackbone=Qwen2.5-Math-7B2026.0291.725.2
HellingerBackbone=Qwen2.5-Math-7B2026.0291.543.8
Total VariationBackbone=Qwen2.5-Math-7B2026.0287.795.4
PearsonBackbone=Qwen2.5-Math-1.5B2026.0286.492.6
Total VariationBackbone=Qwen2.5-Math-1.5B2026.0283.113.8
Reverse KLBackbone=Qwen2.5-Math-1.5B2026.0282.473
KLBackbone=Qwen2.5-Math-1.5B2026.0281.834.4
HellingerBackbone=Qwen2.5-Math-1.5B2026.0281.114.2
Jensen-ShannonBackbone=Qwen2.5-Math-1.5B2026.0278.994.2
GRPOBackbone=Qwen2.5-Math-1.5B2026.0274.265.2
BaseBackbone=Qwen2.5-Math-1.5B2026.0208
BaseBackbone=Qwen2.5-Math-7B2026.0208