Image de couverture de livre

Debunk the Myth of SFT Generalization

Fuente: arXiv

Enregistré dans:

Détails bibliographiques
Auteurs principaux:	Lin, Xiaofeng, Sang, Hejian, Wang, Zhipeng, Zhang, Xuezhou
Format:	Preprint
Publié:	2025
Sujets:	Machine Learning Artificial Intelligence
Accès en ligne:	Acceder al recurso
Tags:	Ajouter un tag Pas de tags, Soyez le premier à ajouter un tag!

Documents similaires

Scaling In-Context Online Learning Capability of LLMs via Cross-Episode Meta-RL
par: Lin, Xiaofeng, et autres
Publié: (2026)

Efficient Reinforcement Learning in Probabilistic Reward Machines
par: Lin, Xiaofeng, et autres
Publié: (2024)

Overconfident Errors Need Stronger Correction: Asymmetric Confidence Penalties for Reinforcement Learning
par: Xu, Yuanda, et autres
Publié: (2026)

PACED: Distillation and On-Policy Self-Distillation at the Frontier of Student Competence
par: Xu, Yuanda, et autres
Publié: (2026)

Not all tokens are needed(NAT): token efficient reinforcement learning
par: Sang, Hejian, et autres
Publié: (2026)

Aligning Diffusion Language Models via Unpaired Preference Optimization
par: Jindal, Vaibhav, et autres
Publié: (2025)

Beyond GRPO and On-Policy Distillation: An Empirical Sparse-to-Dense Reward Principle for Language-Model Post-Training
par: Xu, Yuanda, et autres
Publié: (2026)

TIP: Token Importance in On-Policy Distillation
par: Xu, Yuanda, et autres
Publié: (2026)

Scale-free Adversarial Reinforcement Learning
par: Chen, Mingyu, et autres
Publié: (2024)

State-free Reinforcement Learning
par: Chen, Mingyu, et autres
Publié: (2024)

mSFT: Addressing Dataset Mixtures Overfitting Heterogeneously in Multi-task SFT
par: Koh, Woosung, et autres
Publié: (2026)

Avoiding $\mathbf{exp(R_{max})}$ scaling in RLHF through Preference-based Exploration
par: Chen, Mingyu, et autres
Publié: (2025)

Gradients Must Earn Their Influence: Unifying SFT with Generalized Entropic Objectives
par: Wang, Zecheng, et autres
Publié: (2026)

Memorize Theorems, Not Instances: Probing SFT Generalization through Mathematical Reasoning
par: Peng, Ruiying, et autres
Publié: (2026)

Provably Mitigating Overoptimization in RLHF: Your SFT Loss is Implicitly an Adversarial Regularizer
par: Liu, Zhihan, et autres
Publié: (2024)

RL Fine-Tuning Heals OOD Forgetting in SFT
par: Jin, Hangzhan, et autres
Publié: (2025)

PepThink-R1: LLM for Interpretable Cyclic Peptide Optimization with CoT SFT and Reinforcement Learning
par: Wang, Ruheng, et autres
Publié: (2025)

Quagmires in SFT-RL Post-Training: When High SFT Scores Mislead and What to Use Instead
par: Kang, Feiyang, et autres
Publié: (2025)

Consolidation or Adaptation? PRISM: Disentangling SFT and RL Data via Gradient Concentration
par: Zhao, Yang, et autres
Publié: (2026)

SFT-GO: Supervised Fine-Tuning with Group Optimization for Large Language Models
par: Kim, Gyuhak, et autres
Publié: (2025)

Complementing reinforcement learning with SFT through logit averaging in the post training of LLMs
par: Gan, Xingwei, et autres
Publié: (2026)

GAC: Noise-Aware Adaptive Mixing for Hybrid SFT-RL Post-Training
par: Hu, Yuelin, et autres
Publié: (2026)

Accelerating RL for LLM Reasoning with Optimal Advantage Regression
par: Brantley, Kianté, et autres
Publié: (2025)

Reference-Sampled Boltzmann Projection for KL-Regularized RLVR: Target-Matched Weighted SFT, Finite One-Shot Gaps, and Policy Mirror Descent
par: Shu, Yao, et autres
Publié: (2026)

Implicit Reward as the Bridge: A Unified View of SFT and DPO Connections
par: Wang, Bo, et autres
Publié: (2025)

A Practical Two-Stage Recipe for Mathematical LLMs: Maximizing Accuracy with SFT and Efficiency with Reinforcement Learning
par: Yoshihara, Hiroshi, et autres
Publié: (2025)

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

DynamixSFT: Dynamic Mixture Optimization of Instruction Tuning Collections
par: Shin, Haebin, et autres
Publié: (2025)

SFT-then-RL Outperforms Mixed-Policy Methods for LLM Reasoning
par: Limozin, Alexis, et autres
Publié: (2026)

SFT-GRPO Data Overlap as a Post-Training Hyperparameter for Autoformalization
par: Su, Xiaole, et autres
Publié: (2026)

Harnessing Network Effect for Fake News Mitigation: Selecting Debunkers via Self-Imitation Learning
par: Xu, Xiaofei, et autres
Publié: (2024)

Climbing the Ladder of Reasoning: What LLMs Can-and Still Can't-Solve after SFT?
par: Sun, Yiyou, et autres
Publié: (2025)

RLHF in an SFT Way: From Optimal Solution to Reward-Weighted Alignment
par: Du, Yuhao, et autres
Publié: (2025)

Continual SFT Matches Multimodal RLHF with Negative Supervision
par: Zhu, Ke, et autres
Publié: (2024)

Stable but Wrong: When More Data Degrades Scientific Conclusions
par: Zhang, Zhipeng, et autres
Publié: (2026)

Towards On-Policy SFT: Distribution Discriminant Theory and its Applications in LLM Training
par: Zhang, Miaosen, et autres
Publié: (2026)

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them
par: Rajani, Neel, et autres
Publié: (2025)

Q-SFT: Q-Learning for Language Models via Supervised Fine-Tuning
par: Hong, Joey, et autres
Publié: (2024)

PDC & DM-SFT: A Road for LLM SQL Bug-Fix Enhancing
par: Duan, Yiwen, et autres
Publié: (2024)

SFT Memorizes, RL Generalizes: A Comparative Study of Foundation Model Post-training
par: Chu, Tianzhe, et autres
Publié: (2025)