Supervised Fine-Tuning

Also known as SFT

Supervised fine-tuning trains a pretrained model to follow examples of desired behavior, from answering questions to reasoning, using tools, and interpreting images, audio, or other domain-specific inputs. Common approaches curate or generate instruction and agent-trajectory data, distill teacher outputs, and use methods such as LoRA or selective token and layer updates to control training cost and influence. Current work also adjusts what the model learns through token weighting, self-distillation, and constraints that preserve its original output distribution. The central tension is gaining reliable task performance without losing general capabilities or limiting later learning; how SFT should complement reinforcement learning depends on the data, task, and model.

Latest papers

231 papers29 papers in the last four weeks, up 21% on the four weeks before. 0.2% of all new papers.

Unifying and Beyond the Theory of Pervasive Capability Degradation in LLM Post-training SFT from Feature Interference
Nov 3, 2026Hang Chen, Jiaying Zhu, Wenya WangLanguage Model Post-Training Fine-Tuning Language Models
College of Computing and Data Science, Nanyang Technological University · School of Computer Science and Engineering, The Chinese University of Hong Kong
Self-Privileged Fine-Tuning
Nov 3, 2026Yangrong Hu, Xueyu Zhou, Yue Wu +1Supervised Fine-Tuning Adaptive Loss Weighting
Department of Data Science and Artificial Intelligence The Hong Kong Polytechnic University · Department of Applied Mathematics The Hong Kong Polytechnic University
Correct Labels Are Not Enough: Reliable Metadata Can Silently Control Model Behavior
Nov 3, 2026Renee Jia, Jinzhou WuLLM Reliability Supervised Fine-Tuning
R2M AI · Cornell University
Balancing Chain-of-Thought and Answer Supervision for Small Language Models on Medical Question Answering
Nov 2, 2026Le Zhou, Xiao Luo, Chinmay Chandra +1Fine-Tuning Supervised Fine-Tuning
Pace University, New York City, USA · Southern Methodist University, Dallas, USA
MicroMouse: Masking Harmful Supervision in Long CoT via Relative Potential
Nov 2, 2026Xiaowei Zhu, Hengzhu Tang, Zhenyu Zhang +3CoT Supervision Selective Fine-Tuning
Institute of Information Engineering, Chinese Academy of Sciences, Beijing, China · School of Cyber Security, University of Chinese Academy of Sciences, Beijing, China · Baidu Inc., Beijing, China
EvoMCP: Co-evolving MCP agents and training data
Oct 30, 2026Xuanang Gao, Kepeng Lin, Xinyan Shi +6LLM Agent Training Tool-Using Agents
Shanghai Jiao Tong University · Ant Group · Huazhong University of Science and Technology +1
Can Personal Requirements Become Supervision for Planning?
Oct 29, 2026Junwei Wu, Runze Yan, Yuzhang Xie +5LLM Planning Constraint-Guided Reasoning in Language Models
Emory University, Atlanta, Georgia, USA · Microsoft, San Francisco, California, USA
Scaling Mathematical Proof Supervision from Structured Textbooks
Oct 29, 2026Youqing Fang, Zhouqi Hua, Yinhao Tang +7LLM Mathematical Reasoning Structured Reasoning in Language Models
University of Science and Technology of China · Shanghai AI Laboratory · Fudan University +1
Aligning Large Language Models for Safety in African Languages: A Case Study of Hausa
Oct 27, 2026Valdini Douglace Lemofouet, Blessing Ngozi Uzor, Paula Chikaodinaka Anyanwu +4LLM Safety Alignment Language Model Safety Evaluation
African Institute for Mathematical Sciences (AIMS), Cameroon · Bayero University Kano · University of Hamburg +1
Entropic Control for Homeostatic Optimization: Pre-empting Catastrophic Forgetting via Dynamic Feedback
Oct 25, 2026Shangxi Wu, Xiaorong Yan, Xiaojun Meng +2Catastrophic Forgetting Continual Learning for LLMs
Beijing Jiaotong University · The Hong Kong Polytechnic University · Huawei Noah’s Ark Lab
Intervention-Certified Amplification: A Validity Gate for $σ$ -Weighted Post-Training
Oct 24, 2026Leiyue Chen, Meng Wei, Xuefeng Hu +1RL Post-Training RL for Language Model Reasoning
ByteDance
Trajectory2Gym: Decision-Grounded Compositional Interaction Synthesis for Tool-Use Agents
Oct 24, 2026Yihao Hu, Zhihao Wen, Yitong Lin +5Tool-Using Agents LLM Agent Trajectory Synthesis
Westlake University (westlake.edu) · Ant International (ant-intl.com)
Why Constrained Updates Still Forget: A Functional-Path Analysis of Continual SFT
Oct 21, 2026Qingsen Ma, Dianyun Wang, Jiahong Liu +1Supervised Fine-Tuning Continual Learning for LLMs
The Chinese University of Hong Kong · Beijing University of Posts and Telecommunications
Self-Distillation for Supervised Fine-Tuning
Oct 18, 2026Wenxuan Zhu, Bang An, Shuming Liu +6Supervised Fine-Tuning Language Model Distillation
King Abdullah University of Science and Technology · Shanghai Jiao Tong University
Hidden Cartographer: Hidden State Guided Data Selection for Dual Budgeted Instruction Tuning
Oct 18, 2026Zixuan Qin, Qi Qin, Erbo Li +3Training Data Selection Instruction-Tuning Data Selection
Center for Applied Statistics, School of Statistics, Renmin University of China · Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing
The Ephemeral Cost of Switching Optimizers Across LLM Training Stages
Oct 15, 2026Xinyu Zhao, Davis Wertheimer, Ashutosh Sharma +2LLM Training LLM Post-Training
MIT · MIT-IBM Watson AI Lab
AutoSFT: An Agentic System for Automated Data Research in Supervised Fine-Tuning
Oct 15, 2026Bingxiang He, Pengqi Sun, Shihao Peng +18Supervised Fine-Tuning Instruction-Tuning Data Selection
Tsinghua University · Beijing University of Posts and Telecommunications · ModelBest Inc. +2
Multi-SFT: Permissively-Licensed Multilingual Instruction-Following
Oct 15, 2026Daniela Baiamonte, Kamyar Zeinalipour, Elena Fano +2Instruction Tuning Multilingual Language Models
Yukai! · University of Siena
Learning When to Withdraw: Curriculum Coverage and Rule-Specific Failures in Fine-Tuned Language Agents
Oct 15, 2026Mohamed AbouzahraLLM Agent Evaluation LLM Agent Safety
California State University, Monterey Bay
Beyond Distribution Matching: Selective Margin Preservation for Supervised Fine-Tuning
Oct 14, 2026Yuxuan Wang, Guangxiang Zhao, Xinyu Tan +6Supervised Fine-Tuning Selective Fine-Tuning
Peking University · Xidian University · Qiyuan Tech +1
Post-Training, Finetuning, and Steering:A Field Guide for Curating Training Data
Oct 13, 2026Maria Attarian, Yifan Ruan, Claas Voelcker +5VLA Model Adaptation RL Fine-Tuning
University of Toronto · University of Texas at Austin
Geometric Calibration of Supervised Fine-Tuning
Oct 13, 2026Jianzhu Bao, Jia-Chen Gu, Qingyuan Liu +4LLM Fine-Tuning Supervised Fine-Tuning
Nanyang Technological University · University of California, Los Angeles
Collapse, Repair, and Drift: Dissecting Fine Tuning Dynamics in Speech Language Models
Oct 13, 2026Yichi Wang, Wangjin Zhou, Yizhou Zhang +4Supervised Fine-Tuning Neural Network Training Dynamics
Department of Intelligence Science and Technology, Kyoto University · Xiaowei, WXG, Tencent · Communication University of China
TSRL: Trigger-based Supervised Reinforcement Learning for Continual Agent Training
Oct 13, 2026Yuqian Fu, Kaiwen Jiang, Zijun Song +3Self-Supervised RL Continual Learning for LLM Agents
SKL-MAIS, Institute of Automation, Chinese Academy of Sciences · School of Artificial Intelligence, University of Chinese Academy of Sciences · University of Science and Technology of China
Soft Teacher Forcing
Oct 12, 2026Mohammad Fahes, Cordelia Schmid, Anurag ArnabSupervised Fine-Tuning LLM Reasoning
Google Deepmind
SFT Leaves a Reproducible Signature in Activation Geometry: Calibrating Post-Training Interpretability Against Run-to-Run Noise
Oct 12, 2026Xupeng Chen, Chenqian Le, Jiayi Cheng +2Language Model Post-Training LLM Training
New York University · University of Southern California · Columbia University
Vela: User Behavior as Supervision Sparks LLM Intelligence
Oct 10, 2026Yang Xiao, Yusong Sun, Haoyi Wu +17LLM Training LLM Post-Training
Quality, Not Just Outcome: A Structural Scoring Framework for Selecting Coding-Agent Trajectories for Supervised Fine-Tuning
Oct 9, 2026Sida Li, Yuanxin Liu, Shicheng Li +4AI Coding Agents Training Data Selection
National Key Laboratory for Multimedia Information Processing, Peking University · Huawei Technologies Co., Ltd.
What Should SFT Learn? Conditional Information Gain for Generalizable Post-Training
Oct 9, 2026Yuxuan Gu, Kequan Niu, Xiaocheng Feng +3Supervised Fine-Tuning Language Model Post-Training
Harbin Institute of Technology
Overcoming Prior Barriers: Supervised Fine-Tuning under Long-Tail Distribution
Oct 8, 2026Haohui Wang, Jiahao Xu, Wangzhi Zhan +9Supervised Fine-Tuning Instruction-Tuning Data Selection
Virginia Tech · Amazon · Meta +2

Nov 3, 2026J/K move · Enter open · S save

Unifying and Beyond the Theory of Pervasive Capability Degradation in LLM Post-training SFT from Feature Interference

Hang Chen, Jiaying Zhu, Wenya Wang

College of Computing and Data Science, Nanyang Technological University · School of Computer Science and Engineering, The Chinese University of Hong Kong

Language Model Post-Training Fine-Tuning Language Models Supervised Fine-Tuning

Open paper AskPDF arXiv

Pervasive Capability Degradation (PCD) refers to the phenomenon in which supervised fine-tuning (SFT), while adapting a large language model (LLM) to a target task, often erodes the general capabilities acquired during pre-training. Prevailing explanations attribute PCD to macroscopic distribution shift and therefore lack microscopic manipulability, leaving existing mitigation methods without a complete theoretical foundation. In this paper, we first develop a bottom-up geometric theory of PCD grounded in feature interference under superposition: narrow-domain SFT data disrupt the almost-orthogonal feature structure formed during pre-training, and the resulting interference drives activations across decision hyperplanes in the output Laguerre-Voronoi geometry, flipping predictions on general tasks. This theory unifies existing methods into six paradigms and reveals both the source of their efficacy and their theoretical limits. We then prove that cross-entropy under one-hot supervision cannot resolve PCD on its own, and accordingly propose the AEGIS loss, which directly constrains the drift of activation margins along all competing token directions, thereby decoupling interference suppression from task adaptation. Across 3 model families and 14 SFT tasks, AEGIS outperforms representative methods of every paradigm on target tasks and pervasive capabilities simultaneously, at only about 1.3 times the computational cost of standard SFT. Its generality, scalability, and mechanistic interpretability offer a new perspective on fully resolving PCD.