2 papers
cs.CL2026
GMTS: Gradient Magnitude-based Token Selection Improves RLVR Training for LLM Reasoning
Outongyi Lv, Yuanwei Zhang, Xiaoqun Zhang
Reinforcement learning (RL), particularly RL with Verifiable Rewards (RLVR), has recently emerged as a central paradigm for enhancing large language models' (LLMs) reasoning abilit…
cs.AI2026
Which Tokens Matter? Adaptive Token Selection for RLVR with the Relative Surprisal Index
Outongyi Lv, Yanzhao Zheng, Yuanwei Zhang +5
Reinforcement learning (RL) has become a powerful tool for propelling Large Language Models (LLMs) beyond imitation-based training towards more robust reasoning capabilities. Among…