1
Universal One-third Time Scaling in Learning Peaked Distributions
揭示学习峰值分布时，训练时间与模型性能呈现普适的1/3幂律缩放，ICML 2026前沿理论发现。
arXiv:2602.03685v2 Announce Type: replace Abstract: Training large language models (LLMs) is computationally expensive, partly because the loss exhibi…