Skip to main content

Plot: QKNorm revisited

olmoe17-8x1b-final-eps-noqk uses no QK-Norm but RMSNorm with weights olmoe17-8x1b-final-eps uses non-parametric QK-Norm & RMSNorm with weights
Created on July 6|Last edited on August 29

6k8k10k12k14kStep2.752.82.852.9
1k2k3k4k5kStep46810
Run set
3
Run set 2