ProFit: Leveraging High-Value Signals in SFT via Probability-Guided Token Selection

Benchmark Model Rank Results
arithmetic-reasoning-on-gsm8kProFit (Qwen3-14B-Base)#74Accuracy: 89.62
mathematical-reasoning-on-math-500ProFit (Qwen3-14B-Base)#82Accuracy: 82.85
multiple-choice-question-answering-mcqa-on-gpqa-diamondProFit (Qwen3-14B-Base)#118Accuracy: 46.53
mathematical-reasoning-on-aime24ProFit (Qwen3-14B-Base)#127Acc: 16.56