https://auto-driving-blog.vercel.app/tags/%E8%AE%A1%E7%AE%97%E9%A2%84%E7%AE%97/ Recent content in 计算预算 on Elon's AD Insight https://auto-driving-blog.vercel.app/images/share.png https://auto-driving-blog.vercel.app/images/share.png Hugo zh-cn Sun, 19 Jul 2026 00:00:00 +0000 https://auto-driving-blog.vercel.app/posts/knowledge/scaling-law%E4%B8%8E%E6%95%B0%E6%8D%AE%E7%AD%96%E7%95%A5/ Sun, 19 Jul 2026 00:00:00 +0000 https://auto-driving-blog.vercel.app/posts/knowledge/scaling-law%E4%B8%8E%E6%95%B0%E6%8D%AE%E7%AD%96%E7%95%A5/ 模型越大越好是一个粗糙的直觉，实际工程中需要在计算预算、数据配比、模型容量之间找到最优平衡。本文讲解scaling law的核心公式与迁移规律、数据混合配比的经验法则、训练曲线异常诊断、以及VLA模型训练中的数据预处理Pipeline与curriculum learning策略。