Model TestError Time Baseline 8.96% 51s BatchNorm 8.25% 66s WeightNorm 8.28% 53s LayerNorm 10.49% 72s RMSNorm 8.83% 61s(15%)

Feb-11-2026, 15:57:32 GMT–Neural Information Processing Systems 

Table 2: BLEU curve of LayerNorm and RMSNorm on devset when initialization center is around0.2.