研究人员发现，可验证奖励强化学习（RLVR）中存在一个关键瓶颈，阻碍了LLM推理优化。研究指出，标准硬裁剪方法中的僵化裁剪决策是原因，它丢弃了裁剪阈值附近的宝贵信号。为解决此问题，他们提出了近边界随机恢复（NSR）方法，这是一种简单的修改，可以随机保留这些略微超出边界的token，从而提高各种模型大小和架构的训练稳定性和性能。 AI
影响 提高了LLM推理任务的训练稳定性和性能，有望实现更强大、更具能力的模型。
排序理由 该集群包含一篇学术论文，详细介绍了一种提高LLM训练稳定性方面的新方法。
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2605.22703v1 Announce Type: new Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a central paradigm for scaling LLM reasoning, yet its optimization often suffers from training instability and suboptimal convergence. Through a systematic dissect…
Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a central paradigm for scaling LLM reasoning, yet its optimization often suffers from training instability and suboptimal convergence. Through a systematic dissection of clipping-based GRPO-style objectives, we …