arXiv 2610.06647 · cs.CL · 2026-10-05
LoGRA: Scaling LLM Reinforcement Learning with Low-Rank Gradient Sketches
Shaokun Zhang, Yifan Zhang, Jian Hu, Yueying Li, Hao Zhang, Binfeng Xu, Jan Kautz, Yi Dong
NVIDIA
在最新榜 #139 2026-W41 第 8 名 本站收录于 2026-10-21
一句话要点
LoGRA 用低秩梯度草图压缩 RL 后训练的显存，并按预测 KL 控制步长保持稳定。
LoGRA cuts RL post-training memory with low-rank gradient sketches and keeps training stable with predicted-KL step control.
阅读 arXiv 原文 PDF Hugging Face 讨论
主题与学科
同主题：强化学习
- Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation ▲218
- Self-Retrospection Distillation: Turning Post-hoc Experiences into Prior Foresight ▲189
- Sharpening Tax in Post-Training ▲111
- RULER: Instance-aware Rubric Rewards for SVG Generation ▲104
- TPPO: Tracking-Value PPO For Language Model Reasoning ▲96
数据说明
元数据来自 arXiv API（已核实）；点赞来自 Hugging Face 每日论文；引用来自 Semantic Scholar。最近刷新 2026-10-22。
要点由本站根据摘要自写，不转载摘要原文，请阅读原文。