STARE: Surprisal-Guided Token-Level Advantage Reweighting for Policy Entropy Stability
Open paper
Benchmark
Model
Rank
Results
mathematical-reasoning-on-aime24
STARE-O1 (DeepSeek-R1-Distill-Qwen-1.5B)
#76
Acc: 53.8
mathematical-reasoning-on-aime-2025
STARE-O1 (DeepSeek-R1-Distill-Qwen-1.5B)
#95
Accuracy: 41.5
Rank counts only results with a code link.