ResearchBenchmarksReward Prediction on PIQA (out-of-domain)Follow82.22AccuracyR1-p71.8274.5277.2279.92Mar 13, 2026Evaluation ResultsMethodMethodLinksAccuracyR1-pBackbone=Qwen2.5-3B-In...Backbone=Qwen2.5-3B-Instruct2026.0382.22R1-OracleBackbone=Qwen2.5-3B-In...Backbone=Qwen2.5-3B-Instruct2026.0382MemRewardBackbone=Qwen2.5-3B-In...Backbone=Qwen2.5-3B-Instruct2026.0381.78MemRewardBackbone=Qwen2.5-1.5B-...Backbone=Qwen2.5-1.5B-Instruct2026.0379.33R1-OracleBackbone=Qwen2.5-1.5B-...Backbone=Qwen2.5-1.5B-Instruct2026.0379.11R1-pBackbone=Qwen2.5-1.5B-...Backbone=Qwen2.5-1.5B-Instruct2026.0372.22