LSRIF: Logic-Structured Reinforcement Learning for Instruction Following

Benchmark Model Rank Results
instruction-following-on-mtbenchDistill-Qwen-14B (LsrIF)–Score (1-10): 7.0
instruction-following-on-multichallengeQwen3-8B (LsrIF)–Accuracy: 32.3
mathematical-reasoning-on-aime-2025Distill-Qwen-14B (LsrIF)–Accuracy: 49.6
mathematical-reasoning-on-aime24Distill-Qwen-14B (LsrIF)–Acc: 70.2
multiple-choice-question-answering-mcqa-on-gpqa-diamondDistill-Qwen-14B (LsrIF)–Accuracy: 60.1