ResearchBenchmarksReward Prediction on NuminaMath (out-of-domain)Follow42.22AccuracyMemReward31.133634.011836.8939.7682Mar 13, 2026Evaluation ResultsMethodMethodLinksAccuracyMemRewardBackbone=Qwen2.5-3B-In...Backbone=Qwen2.5-3B-Instruct2026.0342.22R1-OracleBackbone=Qwen2.5-3B-In...Backbone=Qwen2.5-3B-Instruct2026.0339.33R1-pBackbone=Qwen2.5-3B-In...Backbone=Qwen2.5-3B-Instruct2026.0336.44MemRewardBackbone=Qwen2.5-1.5B-...Backbone=Qwen2.5-1.5B-Instruct2026.0334.67R1-OracleBackbone=Qwen2.5-1.5B-...Backbone=Qwen2.5-1.5B-Instruct2026.0332R1-pBackbone=Qwen2.5-1.5B-...Backbone=Qwen2.5-1.5B-Instruct2026.0331.56