Learning to Hint for Reinforcement Learning

Benchmark Model Rank Results
mathematical-reasoning-on-minerva-mathHiLL (Qwen2.5-7B-Instruct)#23Accuracy: 40.1
mathematical-reasoning-on-olympiadbenchHiLL (Qwen2.5-7B-Instruct)#31Accuracy: 48.1
mathematical-reasoning-on-math-500HiLL (Qwen2.5-7B-Instruct)#86Accuracy: 81.8
multi-task-language-understanding-on-mmlu-proHiLL (Qwen2.5-7B-Instruct)#93Accuracy: 61.6
mathematical-reasoning-on-aime24HiLL (Qwen2.5-7B-Instruct)#125Acc: 16.9
multiple-choice-question-answering-mcqa-on-gpqa-diamondHiLL (Qwen2.5-7B-Instruct)#125Accuracy: 40.4
mathematical-reasoning-on-aime-2025HiLL (Qwen2.5-7B-Instruct)#135Accuracy: 15.3