paper-with-me

Papers

Probing RLVR training instability through the lens of objective-level hacking

2026-02-01 · Yiming Dong, Kun Fu, Haoyu Li, Xinyuan Zhu, Yurou Liu, Lijing Shao, Jieping Ye, Zheng Wang arxiv

Prolonged reinforcement learning with verifiable rewards (RLVR) has been shown to drive continuous improvements in the reasoning capabilities of large language models, but the training is often prone to instabilities, especially in Mixture-of-Experts (MoE) architectures. Training instability severely undermines model capability improvement, yet its underlying causes and mechanisms remain poorly understood. In this work, we introduce a principled framework for understanding RLVR instability through the lens of objective-level hacking. Unlike reward hacking, which arises from exploitable verifiers, objective-level hacking emerges from token-level credit misalignment and is manifested as system-level spurious signals in the optimization objective. Grounded in our framework, together with extensive experiments on a 30B MoE model, we trace the origin and formalize the mechanism behind a key pathological training dynamic in MoE models: the abnormal growth of the training-inference discrepancy, a phenomenon widely associated with instability but previously lacking a mechanistic explanation. These findings provide a concrete and causal account of the training dynamics underlying instabilities in MoE models, offering guidance for the design of stable RLVR algorithms.

📄 PDF Abstract BibTeX arXiv:2602.01103

Code (0)

등록된 구현이 없습니다.

Tasks

Reinforcement Learning

Similar Papers 제목 키워드 기반

Understanding Diversity Collapse in RLVR via the Lens of Overtraining

2026-06-13 · Suqin Yuan, Jinkun Chen, Jiyang Zheng, Muyang Li 외 arxiv

Reinforcement learning with verifiable rewards (RLVR) has become a key approach for enhancing the reasoning abilities of large language models. However, RLVR often suffers from \emph{diversity collapse}: Pass@$1$ improve…

Reinforcement Learning

On the Direction of RLVR Updates for LLM Reasoning: Identification and Exploitation

2026-03-23 · Kexin Huang, Haoming Meng, Junkang Wu, Jinda Lu 외 arxiv

Reinforcement learning with verifiable rewards (RLVR) has substantially improved the reasoning capabilities of large language models. While existing analyses identify that RLVR-induced changes are sparse, they primarily …

Reinforcement Learning

TimeLens: Rethinking Video Temporal Grounding with Multimodal LLMs

2025-12-16 · Jun Zhang, Teng Wang, Yuying Ge, Yixiao Ge 외 arxiv

This paper does not introduce a novel method but instead establishes a straightforward, incremental, yet essential baseline for video temporal grounding (VTG), a core capability in video understanding. While multimodal l…

Reinforcement Learning

Geometry-Preserving Orthonormal Initialization for Low-Rank Adaptation in RLVR

2026-06-30 · Ruijia Zhang, Jiacheng Zhu, Hanqing Zhu, Laixi Shi arxiv

Low-rank adaptation (LoRA) and its variants enable parameter-efficient fine-tuning of large language models under the supervised fine-tuning (SFT) paradigm. However, their efficacy and behavior under Reinforcement learni…

parameter-efficient fine-tuningReinforcement LearningMathematical Reasoning

Decouple before Integration: Test-time Synthesis of SFT and RLVR Task Vectors

2026-05-01 · Chaohao Yuan, Chenghao Xiao, Yu Rong, Hong Cheng 외 arxiv

SFT and RLVR represent two fundamental yet distinct paradigms for LLM post-training, each excelling in distinct dimensions. SFT expands knowledge breadth while RLVR enhances reasoning depth. Yet integrating these complem…

Mathematical Reasoning