Reinforcement Fine-Tuning Naturally Mitigates Forgetting in Continual Post-Training

• URL: http://arxiv.org/abs/2507.05386v3
• Date: Tue, 30 Sep 2025 07:32:01 GMT
• Title: Reinforcement Fine-Tuning Naturally Mitigates Forgetting in Continual Post-Training
• Authors: Song Lai, Haohan Zhao, Rong Feng, Changyi Ma, Wenzhuo Liu, Hongbo Zhao, Xi Lin, Dong Yi, Min Xie, Qingfu Zhang, Hongbin Liu, Gaofeng Meng, Fei Zhu,
• Abstract summary: This paper presents a comparative analysis of two core post-training paradigms: supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT)<n>Our experiments are conducted on a benchmark comprising seven diverse multimodal tasks.
• Score: 36.69514399442043
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Continual post-training (CPT) is a popular and effective technique for adapting foundation models like multimodal large language models to specific and ever-evolving downstream tasks. While existing research has primarily concentrated on methods like data replay, model expansion, or parameter regularization, the fundamental role of the learning paradigm within CPT remains largely unexplored. This paper presents a comparative analysis of two core post-training paradigms: supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT), investigating their respective impacts on knowledge retention during CPT. Our experiments are conducted on a benchmark comprising seven diverse multimodal tasks, utilizing Qwen2.5-VL-7B-Instruct as the base model for continual post-training. The investigation yields two significant findings: (1) When continuously learning on downstream tasks, SFT leads to catastrophic forgetting of previously learned tasks. In contrast, RFT inherently preserves prior knowledge and achieve performance comparable to multi-task training. (2) RFT successfully protects and even enhances the model's general knowledge on standard benchmarks (e.g., MMMU and MMLU-Pro). Conversely, SFT degrades general model capabilities severely. Further analysis reveals that this stability is not primarily due to explicit mechanisms like KL penalty or chain-of-thought reasoning. Instead, we identify an implicit regularization mechanism inherent to RFT as a key contributing factor. Our theoretical analysis suggests that RFT's gradient updates are naturally scaled by the reward variance, acting as a data-dependent regularizer that inherently protects previously acquired knowledge. Finally, we propose a rollout-based instance filtering algorithm to enhance the stability and efficiency of RFT. Our comprehensive study demonstrates the superiority of RFT as a robust paradigm for continual post-training.

Related papers

• How and Why LLMs Generalize: A Fine-Grained Analysis of LLM Reasoning from Cognitive Behaviors to Low-Level Patterns [51.02752099869218]
  Large Language Models (LLMs) display strikingly different generalization behaviors.<n>We introduce a novel benchmark that decomposes reasoning into atomic core skills.<n>We show that RL-tuned models maintain more stable behavioral profiles and resist collapse in reasoning skills, whereas SFT models exhibit sharper drift and overfit to surface patterns.
  arXiv  Detail & Related papers  (2025-12-30T08:16:20Z)
• Trust-Region Adaptive Policy Optimization [82.09255251747818]
  Post-training methods play an important role in improving large language models' (LLMs) complex reasoning abilities.<n>We introduce TRAPO, a framework that interleavesSupervised Fine-Tuning (SFT) and Reinforcement Learning (RL) within each training instance.<n>Experiments on five mathematical reasoning benchmarks show that TRAPO consistently surpasses standard SFT, RL, and SFT-then-RL pipelines.
  arXiv  Detail & Related papers  (2025-12-19T14:37:07Z)
• Anchored Supervised Fine-Tuning [26.17356786243252]
  Post-training of large language models involves a trade-off between supervised fine-tuning and reinforcement learning.<n> Dynamic Fine-Tuning (DFT) recently emerged as a promising middle ground, reweighting SFT objectives with token probabilities.<n>We propose Anchored Supervised Fine-Tuning (ASFT) to augment DFT's reweighting with lightweight KL regularization to preserve tightness while ensuring stability.
  arXiv  Detail & Related papers  (2025-09-28T08:58:12Z)
• On-Policy RL Meets Off-Policy Experts: Harmonizing Supervised Fine-Tuning and Reinforcement Learning via Dynamic Weighting [91.38734024438357]
  Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) are two prominent post-training paradigms for refining the capabilities and aligning the behavior of Large Language Models (LLMs)<n>Existing approaches that integrate SFT and RL often face the risk of disrupting established response patterns and inducing overfitting to expert data.<n>We propose CHORD, a framework for Controllable Harmonization of On- and Off-Policy Reinforcement Learning via Dynamic Weighting.
  arXiv  Detail & Related papers  (2025-08-15T11:20:03Z)
• Revisiting Continual Semantic Segmentation with Pre-trained Vision Models [53.56065605992639]
  Continual Semantic (CSS) seeks to incrementally learn to segment novel classes while preserving knowledge of previously encountered ones.<n>Recent advancements in CSS have been driven by the adoption of Pre-trained Vision Models (PVMs) as backbones.<n>Among existing strategies, Direct Fine-Tuning (DFT), which sequentially fine-tunes the model across classes, remains the most straightforward approach.
  arXiv  Detail & Related papers  (2025-08-06T09:51:46Z)
• MLLM-CBench:A Comprehensive Benchmark for Continual Instruction Tuning of Multimodal LLMs with Chain-of-Thought Reasoning Analysis [21.091157331212493]
  Multimodal large language models (MLLMs) require continual instruction tuning during their post-training phase to adapt to the dynamic real-world demands.<n>We introduce textbfMLLM-CTBench, a dataset curating seven challenging tasks from six diverse domains with three contributions.
  arXiv  Detail & Related papers  (2025-07-31T07:49:36Z)
• Blending Supervised and Reinforcement Fine-Tuning with Prefix Sampling [35.64557242726578]
  Prefix-RFT is a hybrid approach that synergizes learning from both demonstration and exploration.<n>It not only surpasses the performance of standalone SFT and RFT but also outperforms parallel mixed-policy RFT methods.
  arXiv  Detail & Related papers  (2025-07-02T13:04:09Z)
• Reinforcement Fine-Tuning Enables MLLMs Learning Novel Tasks Stably [80.36077974826865]
  Post-training algorithms such as Supervised Fine-Tuning (SFT) and Reinforcement Fine-Tuning (RFT) are widely used to adapt multimodal large language models to downstream tasks.<n>We study the behavior of SFT and RFT on an open-source multimodal model, Qwen2.5-VL.<n>Our experiments reveal a sharp trade-off: SFT enables rapid task acquisition but leads to catastrophic forgetting, whereas RFT learns more slowly on novel tasks but maintains prior knowledge.
  arXiv  Detail & Related papers  (2025-06-30T04:15:01Z)
• Implicit Reward as the Bridge: A Unified View of SFT and DPO Connections [65.36449542323277]
  We present a unified theoretical framework bridgingSupervised Fine-Tuning (SFT) and preference learning in Large Language Model (LLM) post-training.<n>We propose a simple yet effective learning rate reduction approach that yields significant performance improvements.
  arXiv  Detail & Related papers  (2025-06-15T05:42:29Z)
• Dual Decomposition of Weights and Singular Value Low Rank Adaptation [9.048461365342204]
  We propose DuDe, a novel approach that decomposes weight matrices into magnitude and direction components.<n>Our evaluation demonstrates DuDe's superior performance and robustness, achieving up to 48.35% accuracy on MMLU and 62.53% ($pm$ 1.59) accuracy on GSM8K.
  arXiv  Detail & Related papers  (2025-05-20T13:49:15Z)
• R-TPT: Improving Adversarial Robustness of Vision-Language Models through Test-Time Prompt Tuning [97.49610356913874]
  We propose a robust test-time prompt tuning (R-TPT) for vision-language models (VLMs)<n>R-TPT mitigates the impact of adversarial attacks during the inference stage.<n>We introduce a plug-and-play reliability-based weighted ensembling strategy to strengthen the defense.
  arXiv  Detail & Related papers  (2025-04-15T13:49:31Z)
• Understanding Fine-tuning in Approximate Unlearning: A Theoretical Perspective [39.958103832214135]
  Fine-tuning (FT) methods have become one of the fundamental approaches for approximating unlearning.<n>We present the first theoretical analysis of FT methods for machine unlearning within a linear regression framework.<n>We propose a novel Retention-Based Masking (RBM) strategy that constructs a weight saliency map based on the remaining dataset.
  arXiv  Detail & Related papers  (2024-10-04T18:01:52Z)
• Doubly Robust Instance-Reweighted Adversarial Training [107.40683655362285]
  We propose a novel doubly-robust instance reweighted adversarial framework. Our importance weights are obtained by optimizing the KL-divergence regularized loss function. Our proposed approach outperforms related state-of-the-art baseline methods in terms of average robust performance.
  arXiv  Detail & Related papers  (2023-08-01T06:16:18Z)
• TWINS: A Fine-Tuning Framework for Improved Transferability of Adversarial Robustness and Generalization [89.54947228958494]
  This paper focuses on the fine-tuning of an adversarially pre-trained model in various classification tasks. We propose a novel statistics-based approach, Two-WIng NormliSation (TWINS) fine-tuning framework. TWINS is shown to be effective on a wide range of image classification datasets in terms of both generalization and robustness.
  arXiv  Detail & Related papers  (2023-03-20T14:12:55Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.