SFT Doesn't Always Hurt General Capabilities: Revisiting Domain-Specific Fine-Tuning in LLMs

• URL: http://arxiv.org/abs/2509.20758v1
• Date: Thu, 25 Sep 2025 05:28:22 GMT
• Title: SFT Doesn't Always Hurt General Capabilities: Revisiting Domain-Specific Fine-Tuning in LLMs
• Authors: Jiacheng Lin, Zhongruo Wang, Kun Qian, Tian Wang, Arvind Srinivasan, Hansi Zeng, Ruochen Jiao, Xie Zhou, Jiri Gesi, Dakuo Wang, Yufan Guo, Kai Zhong, Weiqi Zhang, Sujay Sanghavi, Changyou Chen, Hyokun Yun, Lihong Li,
• Abstract summary: Supervised Fine-Tuning (SFT) is a common approach to adapt Large Language Models (LLMs) to specialized tasks.<n>We show that SFT does not always hurt: using a smaller learning rate can substantially mitigate general performance degradation.
• Score: 53.77646961962239
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Supervised Fine-Tuning (SFT) on domain-specific datasets is a common approach to adapt Large Language Models (LLMs) to specialized tasks but is often believed to degrade their general capabilities. In this work, we revisit this trade-off and present both empirical and theoretical insights. First, we show that SFT does not always hurt: using a smaller learning rate can substantially mitigate general performance degradation while preserving comparable target-domain performance. We then provide a theoretical analysis that explains these phenomena and further motivates a new method, Token-Adaptive Loss Reweighting (TALR). Building on this, and recognizing that smaller learning rates alone do not fully eliminate general-performance degradation in all cases, we evaluate a range of strategies for reducing general capability loss, including L2 regularization, LoRA, model averaging, FLOW, and our proposed TALR. Experimental results demonstrate that while no method completely eliminates the trade-off, TALR consistently outperforms these baselines in balancing domain-specific gains and general capabilities. Finally, we distill our findings into practical guidelines for adapting LLMs to new domains: (i) using a small learning rate to achieve a favorable trade-off, and (ii) when a stronger balance is further desired, adopt TALR as an effective strategy.

Related papers

• Over-Alignment vs Over-Fitting: The Role of Feature Learning Strength in Generalization [8.58740389510812]
  We develop a theoretical analysis of gradient flow dynamics in two-layer ReLU nets trained with logistic loss.<n>An excessively large FLS induces an $textitover-alignment$ phenomenon that degrades generalization, while an overly small FLS leads to $textitover-fitting$.
  arXiv  Detail & Related papers  (2026-01-31T17:43:02Z)
• Data Distribution as a Lever for Guiding Optimizers Toward Superior Generalization in LLMs [60.68927774057402]
  We show, for the first time, that a lower simplicity bias induces a better generalization.<n>Motivated by this insight, we demonstrate that the training data distribution by upsampling or augmenting examples learned later in training similarly reduces SB and leads to improved generalization.<n>Our strategy improves the performance of multiple language models including Phi2-2.7B, Llama3.2-1B, Gemma3-1B-PT, Qwen3-0.6B-Base-achieving relative accuracy gains up to 18% when fine-tuned with AdamW and Muon.
  arXiv  Detail & Related papers  (2026-01-31T07:40:36Z)
• Learning Regularizers: Learning Optimizers that can Regularize [0.959678985347394]
  We show that Learneds (LOs) can be trained to internalize the effects of traditional regularization techniques without explicitly applying them to the objective function.<n>Regularized LOs consistently outperform their unregularized counterparts in terms of test accuracy and generalization.<n>Our results suggest that LOs can inherently learn regularization properties, textitchallenging the conventional necessity of explicit optimizee loss regularization.
  arXiv  Detail & Related papers  (2025-10-10T03:23:28Z)
• T-TAMER: Provably Taming Trade-offs in ML Serving [32.526955555483354]
  We present a general framework, T-Tamer, which formalizes this setting as a multi-stage decision process.<n>Our main result shows that recall is both necessary and sufficient for achieving provable performance guarantees.<n>The results show that recall-based strategies consistently yield efficient accuracy-latency trade-offs.
  arXiv  Detail & Related papers  (2025-09-26T23:08:03Z)
• NDCG-Consistent Softmax Approximation with Accelerated Convergence [67.10365329542365]
  We propose novel loss formulations that align directly with ranking metrics.<n>We integrate the proposed RG losses with the highly efficient Alternating Least Squares (ALS) optimization method.<n> Empirical evaluations on real-world datasets demonstrate that our approach achieves comparable or superior ranking performance.
  arXiv  Detail & Related papers  (2025-06-11T06:59:17Z)
• DGRO: Enhancing LLM Reasoning via Exploration-Exploitation Control and Reward Variance Management [18.953750405635393]
  Decoupled Group Reward Optimization (DGRO) is a general RL algorithm for Large Language Models (LLMs) reasoning.<n>We show that DGRO achieves state-of-the-art performance on the Logic dataset with an average accuracy of 96.9%, and demonstrates strong generalization across mathematical benchmarks.
  arXiv  Detail & Related papers  (2025-05-19T10:44:49Z)
• MoL for LLMs: Dual-Loss Optimization to Enhance Domain Expertise While Preserving General Capabilities [0.0]
  We propose a novel framework, Mixture of Losses (MoL), which decouples optimization objectives for domain-specific and general corpora.<n>Specifically, cross-entropy (CE) loss is applied to domain-corpus to ensure knowledge acquisition, while Kullback-Leibler (KL) divergence aligns general-corpus training with the base model's foundational capabilities.
  arXiv  Detail & Related papers  (2025-05-17T15:12:47Z)
• Mitigating Forgetting in LLM Fine-Tuning via Low-Perplexity Token Learning [65.23593936798662]
  We show that fine-tuning with LLM-generated data improves target task performance and reduces non-target task degradation.<n>This is the first work to provide an empirical explanation based on token perplexity reduction to mitigate catastrophic forgetting in LLMs after fine-tuning.
  arXiv  Detail & Related papers  (2025-01-24T08:18:56Z)
• How to Alleviate Catastrophic Forgetting in LLMs Finetuning? Hierarchical Layer-Wise and Element-Wise Regularization [15.434072331989878]
  Large Language Models (LLMs) exhibit strong general language capabilities.<n>Fine-tuning these models on domain-specific tasks often leads to catastrophic forgetting, where the model overwrites or loses essential knowledge acquired during pretraining.<n>We propose a novel approach to compute the element-wise importance of model parameters crucial for preserving general knowledge during fine-tuning.
  arXiv  Detail & Related papers  (2025-01-23T13:54:53Z)
• A Unified Generalization Analysis of Re-Weighting and Logit-Adjustment for Imbalanced Learning [129.63326990812234]
  We propose a technique named data-dependent contraction to capture how modified losses handle different classes. On top of this technique, a fine-grained generalization bound is established for imbalanced learning, which helps reveal the mystery of re-weighting and logit-adjustment.
  arXiv  Detail & Related papers  (2023-10-07T09:15:08Z)
• Mitigating the Alignment Tax of RLHF [76.4300447532456]
  aligning LLMs under Reinforcement Learning with Human Feedback can lead to forgetting pretrained abilities, also known as the alignment tax. We propose model averaging to maximize alignment performance while incurring minimal alignment tax. We validate HMA's performance across a range of RLHF algorithms over OpenLLaMA-3B and further extend our findings to Mistral-7B.
  arXiv  Detail & Related papers  (2023-09-12T14:16:54Z)
• Learning Invariant Representations and Risks for Semi-supervised Domain Adaptation [109.73983088432364]
  We propose the first method that aims to simultaneously learn invariant representations and risks under the setting of semi-supervised domain adaptation (Semi-DA) We introduce the LIRR algorithm for jointly textbfLearning textbfInvariant textbfRepresentations and textbfRisks.
  arXiv  Detail & Related papers  (2020-10-09T15:42:35Z)

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.