Memorize Theorems, Not Instances: Probing SFT Generalization through Mathematical Reasoning
About
Supervised Fine-Tuning (SFT) is widely used for task-specific adaptation, yet recent work shows it systematically undermines reasoning generalization. We argue the root cause is not memorization itself, but its target: vanilla SFT drives models to exploit and memorize spurious surface correlations in problem-solution pairs, leaving them brittle to superficial input variations. To address this, we propose Theorem-SFT, which reorients supervision toward explicit theorem application by teaching models how rules are invoked rather than what answers look like. Theorem-SFT yields consistent gains across benchmarks and model families: +8.8% on MATH (LLaMA3.2-3B-Instruct) and +20.27% on GeoQA (Qwen2.5-VL-7B-Instruct) without modality-specific re-training. Fine-tuning MLP layers alone matches full-layers performance, implicating feed-forward components as the primary locus of reasoning rules. Our findings reframe the debate: Generalization failures stem not from memorization as a mechanism, but from memorizing the wrong inductive targets.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Mathematical Reasoning | AGIEval MATH | Accuracy45.6 | 99 | |
| Mathematical Reasoning | Omni-MATH | Accuracy18.4 | 23 | |
| Geometric Reasoning | GeoQA | Accuracy69.73 | 15 | |
| Mathematical Reasoning | GSM1K | Accuracy76.1 | 12 | |
| Mathematical Reasoning | LiveBench | Accuracy53.6 | 12 |