Stabilizing LLM Supervised Fine-Tuning via Explicit Distributional Control
About
Post-training large language models (LLMs) often suffers from catastrophic forgetting, where improvements on a target objective degrade previously acquired capabilities. Recent evidence suggests that this phenomenon is primarily driven by excessive distributional drift during optimization. Motivated by this perspective, we propose Anchored Learning, a simple framework that explicitly controls distributional updates during offline fine-tuning via a dynamically evolving moving anchor. Instead of matching a fixed reference distribution, the anchor interpolates between the current model and a frozen reference to construct an intermediate target that the model distills toward, transforming global fine-tuning into a sequence of local trust-region updates in distribution space. Theoretically, we prove this anchor-based update admits a linear KL-divergence upper bound per iteration, ensuring a stable transition between model distributions. Extensive experiments on iGSM, MedCalc, and IFEval show that Anchored Learning consistently lies on the Pareto frontier of gain-stability trade-offs, achieving near-optimal performance improvements while substantially reducing degradation compared to strong baselines. For example, while standard SFT suffers from over 53% performance degradation on iGSM and MedCalc, Anchored Learning slashes this drop to under 5% while maintaining near-optimal gains (e.g., 75.2% on iGSM).
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Instruction Following | IFEval | IFEval Accuracy60 | 836 | |
| Mathematical Reasoning | Countdown | Accuracy18.4 | 252 | |
| Coding | HumanEval | Pass@170.7 | 168 | |
| Coding | HumanEval+ | Pass@166.5 | 164 | |
| Coding | MBPP | Accuracy58.7 | 145 | |
| Coding | MBPP+ | Pass@161.1 | 117 | |
| Instruction Following | IFEval (test) | -- | 88 | |
| Mathematical Reasoning | COUNTDOWN (test) | Accuracy17 | 84 | |
| Coding | MBPP | Pass@1 Accuracy70.4 | 78 | |
| Coding | HumanEval | Accuracy57.3 | 60 |