{"id":"arxiv/2606.00539","url":"https://arxiv.org/abs/2606.00539","tags":["cs.LG","math.OC","stat.ML"],"score":0,"title":"GNMR: Runtime Stability Control for Low-Precision Large Language Model Training","domain":"cs.LG","source":"arxiv","authors":["Boao Kong","Weichen Jia","Engao Zhang","Guohong Li","Yonghan Dong","Yao Wang","Yaoyuan Wang","Yunke Peng","Kun Yuan"],"reviews":0,"subject":"cs.LG","summary":"Training stability is a key bottleneck in low-precision language model training: efficient low-cost paths can still produce short-lived numerical risks at a small set of operators. We formulate this as runtime stability control and present Gradient Norm-to-Mean Ratio (GNMR), a lightweight controller that compares each recoverable unit's current gradient norm with its historical mean. Together with $Δ$-GNMR for abrupt short-window increases, GNMR maps local risk signals to bounded recovery actions under a hard $\\mathrm{maxO}$ budget and a short lock interval, without changing the numerical…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-05-30","reactions":{}}