Cover Image

Bridging SFT and RL: Dynamic Policy Optimization for Robust Reasoning

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Zhu, Taojie, Xu, Dongyang, Zou, Ding, Zhao, Sen, Hao, Qiaobo, Yang, Zhiguo, He, Yonghong
Format:	Preprint
Published:	2026
Subjects:	Machine Learning
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

BREAD: Branched Rollouts from Expert Anchors Bridge SFT & RL for Reasoning
by: Zhang, Xuechen, et al.
Published: (2025)

SFT-then-RL Outperforms Mixed-Policy Methods for LLM Reasoning
by: Limozin, Alexis, et al.
Published: (2026)

Patch the Distribution Mismatch: RL Rewriting Agent for Stable Off-Policy SFT
by: Wang, Jiacheng, et al.
Published: (2026)

Consolidation or Adaptation? PRISM: Disentangling SFT and RL Data via Gradient Concentration
by: Zhao, Yang, et al.
Published: (2026)

AceReason-Nemotron 1.1: Advancing Math and Code Reasoning through SFT and RL Synergy
by: Liu, Zihan, et al.
Published: (2025)

Loss Gradient Gaussian Width based Generalization and Optimization Guarantees
by: Banerjee, Arindam, et al.
Published: (2024)

RL Fine-Tuning Heals OOD Forgetting in SFT
by: Jin, Hangzhan, et al.
Published: (2025)

Fat-to-Thin Policy Optimization: Offline RL with Sparse Policies
by: Zhu, Lingwei, et al.
Published: (2025)

Light-R1: Curriculum SFT, DPO and RL for Long COT from Scratch and Beyond
by: Wen, Liang, et al.
Published: (2025)

Metis-RISE: RL Incentivizes and SFT Enhances Multimodal Reasoning Model Learning
by: Qiu, Haibo, et al.
Published: (2025)

Post-Training is About States, Not Tokens: A State Distribution View of SFT, RL, and On-Policy Distillation
by: Nie, Dong
Published: (2026)

EdgeInfinite-Instruct: Bridging SFT-Based Optimization and NPU-Level Efficiency for Edge Devices
by: Chen, Jiyu, et al.
Published: (2025)

PDC & DM-SFT: A Road for LLM SQL Bug-Fix Enhancing
by: Duan, Yiwen, et al.
Published: (2024)

RL makes MLLMs see better than SFT
by: Song, Junha, et al.
Published: (2025)

Quagmires in SFT-RL Post-Training: When High SFT Scores Mislead and What to Use Instead
by: Kang, Feiyang, et al.
Published: (2025)

LogicPuzzleRL: Cultivating Robust Mathematical Reasoning in LLMs via Reinforcement Learning
by: Wong, Zhen Hao, et al.
Published: (2025)

Sketched Adaptive Federated Deep Learning: A Sharp Convergence Analysis
by: Chen, Zhijie, et al.
Published: (2024)

Crafting Reversible SFT Behaviors in Large Language Models
by: Lin, Yuping, et al.
Published: (2026)

ReSkill: Reconciling Skill Creation with Policy Optimization in Agentic RL
by: He, Zelin, et al.
Published: (2026)

TMS: Trajectory-Mixed Supervision for Reward-Free, On-Policy SFT
by: Khan, Rana Muhammad Shahroz, et al.
Published: (2026)

GAC: Noise-Aware Adaptive Mixing for Hybrid SFT-RL Post-Training
by: Hu, Yuelin, et al.
Published: (2026)

DynamixSFT: Dynamic Mixture Optimization of Instruction Tuning Collections
by: Shin, Haebin, et al.
Published: (2025)

On the Generalization of SFT: A Reinforcement Learning Perspective with Reward Rectification
by: Wu, Yongliang, et al.
Published: (2025)

RePO: Bridging On-Policy Learning and Off-Policy Knowledge through Rephrasing Policy Optimization
by: Xia, Linxuan, et al.
Published: (2026)

Beyond Scaling Law: A Data-Efficient Distillation Framework for Reasoning
by: Wu, Xiaojun, et al.
Published: (2025)

Structured Role-Aware Policy Optimization for Multimodal Reasoning
by: Jiang, Bingqing, et al.
Published: (2026)

Memorize Theorems, Not Instances: Probing SFT Generalization through Mathematical Reasoning
by: Peng, Ruiying, et al.
Published: (2026)

Dynamics-Predictive Sampling for Active RL Finetuning of Large Reasoning Models
by: Mao, Yixiu, et al.
Published: (2026)

3D-Mol: A Novel Contrastive Learning Framework for Molecular Property Prediction with 3D Information
by: Kuang, Taojie, et al.
Published: (2023)

Decoupling KL and Trajectories: A Unified Perspective for SFT, DAgger, Offline RL, and OPD in LLM Distillation
by: Zhao, Anhao, et al.
Published: (2026)

LLMs Can Learn to Reason Via Off-Policy RL
by: Ritter, Daniel, et al.
Published: (2026)

Robust Post-Training for Generative Recommenders: Why Exponential Reward-Weighted SFT Outperforms RLHF
by: Chidambaram, Keertana, et al.
Published: (2026)

Reshaping Reasoning in LLMs: A Theoretical Analysis of RL Training Dynamics through Pattern Selection
by: Chen, Xingwu, et al.
Published: (2025)

Robust Policy Expansion for Offline-to-Online RL under Diverse Data Corruption
by: He, Longxiang, et al.
Published: (2025)

SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training
by: Chu, Tianzhe, et al.
Published: (2025)

Mechanistic origins of catastrophic forgetting: why RL preserves circuits better than SFT?
by: Nunez, Jeanmely Rojas, et al.
Published: (2026)

Differentially Private Post-Processing for Fair Regression
by: Xian, Ruicheng, et al.
Published: (2024)

Implicit Reward as the Bridge: A Unified View of SFT and DPO Connections
by: Wang, Bo, et al.
Published: (2025)

FAPO: Flawed-Aware Policy Optimization for Efficient and Reliable Reasoning
by: Ding, Yuyang, et al.
Published: (2025)

PVPO: Pre-Estimated Value-Based Policy Optimization for Agentic Reasoning
by: Feng, Wenfeng, et al.
Published: (2025)