# Improved Supervised Fine-Tuning for Large Language Models to Mitigate Catastrophic Forgetting

arXiv: https://arxiv.org/abs/2506.09428 · HTML: https://mlsanity.com/paper/2506.09428/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- Builds on 16 of its 25 references that are in the map.
- First posted on arXiv 11 Jun 2025.

## Builds on (most cited first, 16 of 16)

- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [GPQA: A Graduate-Level Google-Proof Q&A Benchmark](https://mlsanity.com/paper/2311.12022/) (2023)
- [MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark](https://mlsanity.com/paper/2406.01574/) (2024)
- [Continual Learning with Deep Generative Replay](https://mlsanity.com/paper/1705.08690/) (2017)
- [Experience Replay for Continual Learning](https://mlsanity.com/paper/1811.11682/) (2018)
- [SimPO: Simple Preference Optimization with a Reference-Free Reward](https://mlsanity.com/paper/2405.14734/) (2024)
- [Enhancing Chat Language Models by Scaling High-quality Instructional Conversations](https://mlsanity.com/paper/2305.14233/) (2023)
- [The RefinedWeb Dataset for Falcon LLM: Outperforming Curated Corpora with Web Data, and Web Data Only](https://mlsanity.com/paper/2306.01116/) (2023)
- [WildChat: 1M ChatGPT Interaction Logs in the Wild](https://mlsanity.com/paper/2405.01470/) (2024)
- [Mitigating Catastrophic Forgetting in Large Language Models with Self-Synthesized Rehearsal](https://mlsanity.com/paper/2403.01244/) (2024)
- [Analyzing and Reducing Catastrophic Forgetting in Parameter Efficient Tuning](https://mlsanity.com/paper/2402.18865/) (2024)
- [Continual Learning for Natural Language Generation in Task-oriented Dialog Systems](https://mlsanity.com/paper/2010.00910/) (2020)
- [GenQA: Generating Millions of Instructions from a Handful of Prompts](https://mlsanity.com/paper/2406.10323/) (2024)
- [Continual Learning for Task-oriented Dialogue System with Iterative Network Pruning, Expanding and Masking](https://mlsanity.com/paper/2107.08173/) (2021)
- [Sequential Reptile: Inter-Task Gradient Alignment for Multilingual Learning](https://mlsanity.com/paper/2110.02600/) (2021)

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

- [Emergent Slow Thinking in LLMs as Inverse Tree Freezing](https://mlsanity.com/paper/2509.23629/) (2025)
- [RISER: Orchestrating Latent Reasoning Skills for Adaptive Activation Steering](https://mlsanity.com/paper/2601.09269/) (2026)
- [MemSFT: Mitigating Alignment Tax with an External Parametric Memory](https://mlsanity.com/paper/2607.25614/) (2026)
- [Talking to Yourself: Defying Forgetting in Large Language Models](https://mlsanity.com/paper/2602.20162/) (2026)
- [Light Alignment Improves LLM Safety via Model Self-Reflection with a Single Neuron](https://mlsanity.com/paper/2602.02027/) (2026)
- [Crafting Reversible SFT Behaviors in Large Language Models](https://mlsanity.com/paper/2605.06632/) (2026)
- [Granite.Trust Policy Tools: Shareable, Actionable Policies for Generative AI Applications](https://mlsanity.com/paper/2608.23870/) (2026)
- [From Alignment to Access Control: A Framework for GenAI Policy Enforcement](https://mlsanity.com/paper/2609.26682/) (2026)

## Abstract (arXiv)

Supervised Fine-Tuning (SFT) is a critical step for enhancing the instruction-following capabilities of Large Language Models (LLMs) and adapting them to specialized domains. However, SFT often leads to a degradation of the model's general abilities, a phenomenon known as catastrophic forgetting. This problem is exacerbated when third-party practitioners fine-tune open-source models, as the original SFT data is typically not available. To address this challenge, we propose a novel and cost-effective SFT method that effectively mitigates catastrophic forgetting without requiring access to the original SFT data. Our approach first reconstructs the likely instruction distribution of the base model. It then employs a multi-model generation and filtering pipeline to synthesize a high-quality general-purpose dataset. This synthetic dataset is mixed with new, domain-specific data for fine-tuning. Experimental results show that our method not only preserves the model's capabilities in general domains but also improves task-specific performance, outperforming baselines that use publicly available SFT datasets.