Copertina

GAC: Noise-Aware Adaptive Mixing for Hybrid SFT-RL Post-Training

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Hu, Yuelin, Yu, Zhenbo, Cheng, Zhengxue, Liu, Wei, Song, Li
Natura:	Preprint
Pubblicazione:	2026
Soggetti:	Machine Learning Artificial Intelligence
Accesso online:	Acceder al recurso
Tags:	Aggiungi Tag Nessun Tag, puoi essere il primo ad aggiungerne!!

Documenti analoghi

Entropy-Gated Selective Policy Optimization:Token-Level Gradient Allocation for Hybrid Training of Large Language Models
di: Hu, Yuelin, et al.
Pubblicazione: (2026)

Hidden Failure Modes of Gradient Modification under Adam in Continual Learning, and Adaptive Decoupled Moment Routing as a Repair
di: Hu, Yuelin, et al.
Pubblicazione: (2026)

Maximizing Rollout Informativeness under a Fixed Budget: A Submodular View of Tree Search for Tool-Use Agentic Reinforcement Learning
di: Hu, Yuelin, et al.
Pubblicazione: (2026)

GAC: Stabilizing Asynchronous RL Training for LLMs via Gradient Alignment Control
di: Xu, Haofeng, et al.
Pubblicazione: (2026)

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

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

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

AuditRepairBench: A Paired-Execution Trace Corpus for Evaluator-Channel Ranking Instability in Agent Repair
di: Hu, Yuelin, et al.
Pubblicazione: (2026)

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

Bootstrapped Mixed Rewards for RL Post-Training: Injecting Canonical Action Order
di: Gupta, Prakhar, et al.
Pubblicazione: (2025)

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

DAQ: Delta-Aware Quantization for Post-Training LLM Weight Compression
di: Yu, Xiaoming, et al.
Pubblicazione: (2026)

Trust the Batch, On- or Off-Policy: Adaptive Policy Optimization for RL Post-Training
di: Fakoor, Rasool, et al.
Pubblicazione: (2026)

First SFT, Second RL, Third UPT: Continual Improving Multi-Modal LLM Reasoning via Unsupervised Post-Training
di: Wei, Lai, et al.
Pubblicazione: (2025)

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

QaRL: Rollout-Aligned Quantization-Aware RL for Fast and Stable Training under Training--Inference Mismatch
di: Gu, Hao, et al.
Pubblicazione: (2026)

SFT-GRPO Data Overlap as a Post-Training Hyperparameter for Autoformalization
di: Su, Xiaole, et al.
Pubblicazione: (2026)

mSFT: Addressing Dataset Mixtures Overfitting Heterogeneously in Multi-task SFT
di: Koh, Woosung, et al.
Pubblicazione: (2026)

Imbalanced Gradients in RL Post-Training of Multi-Task LLMs
di: Wu, Runzhe, et al.
Pubblicazione: (2025)

AsyncFlow: An Asynchronous Streaming RL Framework for Efficient LLM Post-Training
di: Han, Zhenyu, et al.
Pubblicazione: (2025)

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

SortedRL: Accelerating RL Training for LLMs through Online Length-Aware Scheduling
di: Zhang, Yiqi, et al.
Pubblicazione: (2026)

DiRL: An Efficient Post-Training Framework for Diffusion Language Models
di: Zhu, Ying, et al.
Pubblicazione: (2025)

QuantumNAT: Quantum Noise-Aware Training with Noise Injection, Quantization and Normalization
di: Wang, Hanrui, et al.
Pubblicazione: (2021)

Scale When Needed: Adaptive Neuron-level Mixed Precision Quantization Aware Training
di: Varshney, Ayush K., et al.
Pubblicazione: (2026)

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

Dr. Post-Training: A Data Regularization Perspective on LLM Post-Training
di: Hu, Pingbang, et al.
Pubblicazione: (2026)

Learning-Zone Energy: Online Data Selection for Efficient RL Post-Training
di: Cui, Peng, et al.
Pubblicazione: (2026)

CoScale-RL: Efficient Post-Training by Co-Scaling Data and Computation
di: Chen, Yutong, et al.
Pubblicazione: (2026)

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

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

Debunk the Myth of SFT Generalization
di: Lin, Xiaofeng, et al.
Pubblicazione: (2025)

How to Compress KV Cache in RL Post-Training? Shadow Mask Distillation for Memory-Efficient Alignment
di: Zhu, Rui, et al.
Pubblicazione: (2026)

On Effectiveness and Efficiency of Agentic Tool-calling and RL Training
di: Liu, Tong, et al.
Pubblicazione: (2026)

MDGMIX: Boundary-Aware Subgraph Mixing for Multi-Domain Graph Pre-Training
di: Zheng, Ziyu, et al.
Pubblicazione: (2026)

Role-Based Fault Tolerance System for LLM RL Post-Training
di: Chen, Zhenqian, et al.
Pubblicazione: (2025)

LaRA: Layer-wise Representation Analysis for Detecting Data Contamination in RL Post-Training
di: Gwak, Minju, et al.
Pubblicazione: (2026)

Hybrid-LoRA: Bridging Full Fine-Tuning and Low-Rank Adaptation for Post-Training
di: Zhang, Chengqian, et al.
Pubblicazione: (2026)

AIS: Adaptive Importance Sampling for Quantized RL
di: Zhou, Jiajun, et al.
Pubblicazione: (2026)

SFT-GO: Supervised Fine-Tuning with Group Optimization for Large Language Models
di: Kim, Gyuhak, et al.
Pubblicazione: (2025)