# Debunk the Myth of SFT Generalization

arXiv: https://arxiv.org/abs/2510.00237 · HTML: https://mlsanity.com/paper/2510.00237/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- Placed in RL training of language and agent models, one of 15 directions in Deep reinforcement learning methods.
- Builds on 19 of its 20 references that are in the map.
- First posted on arXiv 30 Sep 2025.

## Placed in

- Deep reinforcement learning methods / [RL training of language and agent models](https://mlsanity.com/deeprl/rl-training-of-language-and-agent-models/)

## Builds on (most cited first, 19 of 19)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Self-Instruct: Aligning Language Models with Self-Generated Instructions](https://mlsanity.com/paper/2212.10560/) (2022)
- [SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training](https://mlsanity.com/paper/2501.17161/) (2025)
- [RAGEN: Understanding Self-Evolution in LLM Agents via Multi-Turn Reinforcement Learning](https://mlsanity.com/paper/2504.20073/) (2025)
- [RL's Razor: Why Online Reinforcement Learning Forgets Less](https://mlsanity.com/paper/2509.04259/) (2025)
- [On the Generalization of SFT: A Reinforcement Learning Perspective with Reward Rectification](https://mlsanity.com/paper/2508.05629/) (2025)
- [On Memorization of Large Language Models in Logical Reasoning](https://mlsanity.com/paper/2410.23123/) (2024)
- [Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning](https://mlsanity.com/paper/2507.00432/) (2025)
- [MATH-Perturb: Benchmarking LLMs' Math Reasoning Abilities against Hard Perturbations](https://mlsanity.com/paper/2502.06453/) (2025)
- [Preserving Diversity in Supervised Fine-Tuning of Large Language Models](https://mlsanity.com/paper/2408.16673/) (2024)
- [Supervised Fine Tuning on Curated Data is Reinforcement Learning (and can be improved)](https://mlsanity.com/paper/2507.12856/) (2025)
- [RL Fine-Tuning Heals OOD Forgetting in SFT](https://mlsanity.com/paper/2509.12235/) (2025)
- [Rethinking Conventional Wisdom in Machine Learning: From Generalization to Scaling](https://mlsanity.com/paper/2409.15156/) (2024)
- [Proximal Supervised Fine-Tuning](https://mlsanity.com/paper/2508.17784/) (2025)
- [Why Distillation can Outperform Zero-RL: The Role of Flexible Reasoning](https://mlsanity.com/paper/2505.21067/) (2025)

## Built on by (most cited first, 8 of 8)

- [Rethinking Generalization in Reasoning SFT: A Conditional Analysis on Optimization, Data, and Model Capability](https://mlsanity.com/paper/2604.06628/) (2026)
- [Towards On-Policy SFT: Distribution Discriminant Theory and its Applications in LLM Training](https://mlsanity.com/paper/2602.12222/) (2026)
- [GRACE: Reinforcement Learning for Grounded Response and Abstention under Contextual Evidence](https://mlsanity.com/paper/2601.04525/) (2026)
- [SED-SFT: Selectively Encouraging Diversity in Supervised Fine-Tuning](https://mlsanity.com/paper/2602.07464/) (2026)
- [KOCO-BENCH: Can Large Language Models Leverage Domain Knowledge in Software Development?](https://mlsanity.com/paper/2601.13240/) (2026)
- [Zipping the Thought: When and How Compressed Reasoning Data Works in LLM Post-Training](https://mlsanity.com/paper/2605.28008/) (2026)
- [When Perplexity Lies: Generation-Focused Distillation of Hybrid Sequence Models](https://mlsanity.com/paper/2603.26556/) (2026)
- [RAISE: Reinforcing Access Control Policy Synthesis in LLMs via Symbolic Evaluation](https://mlsanity.com/paper/2609.33796/) (2026)

## Abstract (arXiv)

A prevailing view holds that supervised fine-tuning (SFT) memorizes training data and fails to generalize, whereas reinforcement learning (RL) attains broader robustness. We revisit this claim through a systematic evaluation on two decision-making benchmarks, Sokoban and General Points, and arrive at a different conclusion. We show that much of SFT's perceived failure stems from frozen-prompt artifacts: when trained on fixed instruction templates, SFT models cling to training semantics rather than adapting to new ones. Introducing prompt diversity during training breaks this shortcut and yields strong generalization to unseen instruction variants without harming in-distribution performance. Beyond instruction shifts, we ask whether SFT can generalize to strictly harder tasks. Here, chain-of-thought (CoT) supervision provides an algorithmic scaffold that markedly improves transfer to more difficult regimes, such as larger Sokoban grids with additional boxes and arithmetic with out-of-distribution values or five-card compositions that increase combinatorial complexity. Finally, combining prompt diversity with CoT achieves the best of both worlds: robust generalization across both instruction-variant and difficulty-variant settings, matching or surpassing RL baselines on our benchmarks while retaining SFT's simplicity and stability. These findings challenge the narrative that SFT is inherently inferior to RL and support a data-centric perspective: with appropriately curated demonstrations, vanilla SFT can generalize as strongly as RL. Code reproducing the results in the paper can be found at: https://github.com/XiaofengLin7/debunking-sft-generalization.