Copertina

Beyond Precision: Training-Inference Mismatch is an Optimization Problem and Simple LR Scheduling Fixes It

Fuente: arXiv

Salvato in:

Dettagli Bibliografici
Autori principali:	Zhang, Yaxiang, Li, Yingru, Liu, Jiacai, Xu, Jiawei, Li, Ziniu, Liu, Qian, Li, Haoyuan
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

A Note on Hybrid Online Reinforcement and Imitation Learning for LLMs: Formulations and Algorithms
di: Li, Yingru, et al.
Pubblicazione: (2025)

Trust Region Masking for Long-Horizon LLM Reinforcement Learning
di: Li, Yingru, et al.
Pubblicazione: (2025)

The Optimal Token Baseline: Variance Reduction for Long-Horizon LLM-RL
di: Li, Yingru, et al.
Pubblicazione: (2026)

Dynamic Vocabulary Pruning: Stable LLM-RL by Taming the Tail
di: Li, Yingru, et al.
Pubblicazione: (2025)

Logit Dynamics in Softmax Policy Gradient Methods
di: Li, Yingru
Pubblicazione: (2025)

Revisiting On-Policy Distillation: Empirical Failure Modes and Simple Fixes
di: Fu, Yuqian, et al.
Pubblicazione: (2026)

Dr. Kernel: Reinforcement Learning Done Right for Triton Kernel Generations
di: Liu, Wei, et al.
Pubblicazione: (2026)

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

Q-Star Meets Scalable Posterior Sampling: Bridging Theory and Practice via HyperAgent
di: Li, Yingru, et al.
Pubblicazione: (2024)

Divergence-Augmented Policy Optimization
di: Wang, Qing, et al.
Pubblicazione: (2025)

Defeating the Training-Inference Mismatch via FP16
di: Qi, Penghui, et al.
Pubblicazione: (2025)

Scalable Exploration via Ensemble++
di: Li, Yingru, et al.
Pubblicazione: (2024)

Diagnosing Training Inference Mismatch in LLM Reinforcement Learning
di: Zhong, Tianle, et al.
Pubblicazione: (2026)

Balanced Aggregation: Understanding and Fixing Aggregation Bias in GRPO
di: Zeng, Zhiyuan, et al.
Pubblicazione: (2026)

Beyond Token-level Supervision: Unlocking the Potential of Decoding-based Regression via Reinforcement Learning
di: Chen, Ming, et al.
Pubblicazione: (2025)

Prior-dependent analysis of posterior sampling reinforcement learning with function approximation
di: Li, Yingru, et al.
Pubblicazione: (2024)

Mitigating Mismatch within Reference-based Preference Optimization
di: Yuan, Suqin, et al.
Pubblicazione: (2026)

Stepwise Guided Policy Optimization: Coloring your Incorrect Reasoning in GRPO
di: Chen, Peter, et al.
Pubblicazione: (2025)

Beyond What to Select: A Plug-and-play Oscillatory Data-Volume Scheduling for Efficient Model Training
di: Yang, Suorong, et al.
Pubblicazione: (2026)

Beyond Training: Optimizing Reinforcement Learning Based Job Shop Scheduling Through Adaptive Action Sampling
di: de Puiseau, Constantin Waubert, et al.
Pubblicazione: (2024)

GreedySnake: Accelerating SSD-Offloaded LLM Training with Efficient Scheduling and Optimizer Step Overlapping
di: Yin, Yishu, et al.
Pubblicazione: (2025)

Decouple Graph Neural Networks: Train Multiple Simple GNNs Simultaneously Instead of One
di: Zhang, Hongyuan, et al.
Pubblicazione: (2023)

Dataset Distillation for Offline Reinforcement Learning
di: Light, Jonathan, et al.
Pubblicazione: (2024)

GradientStabilizer:Fix the Norm, Not the Gradient
di: Huang, Tianjin, et al.
Pubblicazione: (2025)

Early Quantization Shrinks Codebook: A Simple Fix for Diversity-Preserving Tokenization
di: Zhao, Wenhao, et al.
Pubblicazione: (2026)

TEON: Tensorized Orthonormalization Beyond Layer-Wise Muon for Large Language Model Pre-Training
di: Zhang, Ruijie, et al.
Pubblicazione: (2026)

Beyond Fixed Variables: Expanding-variate Time Series Forecasting via Flat Scheme and Spatio-temporal Focal Learning
di: Ma, Minbo, et al.
Pubblicazione: (2025)

Beyond Experience Retrieval: Learning to Generate Utility-Optimized Structured Experience for Frozen LLMs
di: Li, Xuancheng, et al.
Pubblicazione: (2026)

Semi-Clairvoyant Scheduling of Speculative Decoding Requests to Minimize LLM Inference Latency
di: Li, Ruixiao, et al.
Pubblicazione: (2025)

A Simple "Try Again" Can Elicit Multi-Turn LLM Reasoning
di: Liu, Licheng, et al.
Pubblicazione: (2025)

Adaptive Batch-Wise Sample Scheduling for Direct Preference Optimization
di: Huang, Zixuan, et al.
Pubblicazione: (2025)

A Triple-Inertial Accelerated Alternating Optimization Method for Deep Learning Training
di: Yan, Chengcheng, et al.
Pubblicazione: (2025)

Q-Newton: Hybrid Quantum-Classical Scheduling for Accelerating Neural Network Training with Newton's Gradient Descent
di: Li, Pingzhi, et al.
Pubblicazione: (2024)

Controllable Edge-Type-Specific Interpretation in Multi-Relational Graph Neural Networks for Drug Response Prediction
di: Li, Xiaodi, et al.
Pubblicazione: (2024)

PAHQ: Accelerating Automated Circuit Discovery through Mixed-Precision Inference Optimization
di: Wang, Xinhai, et al.
Pubblicazione: (2025)

CAS-Spec: Cascade Adaptive Self-Speculative Decoding for On-the-Fly Lossless Inference Acceleration of LLMs
di: Ning, Zhiyuan, et al.
Pubblicazione: (2025)

Anytime Training with Schedule-Free Spectral Optimization
di: Apte, Anuj, et al.
Pubblicazione: (2026)

Unleashing the Denoising Capability of Diffusion Prior for Solving Inverse Problems
di: Zhang, Jiawei, et al.
Pubblicazione: (2024)

Cross-Domain Policy Adaptation by Capturing Representation Mismatch
di: Lyu, Jiafei, et al.
Pubblicazione: (2024)

ChatScene: Knowledge-Enabled Safety-Critical Scenario Generation for Autonomous Vehicles
di: Zhang, Jiawei, et al.
Pubblicazione: (2024)