rmsnorm-llama-alternative-to-layernorm

Summary

Llama uses RMSNorm in place of the more standard LayerNorm because it drops the mean-centering step, making the per-layer computation slightly lighter. This is a small efficiency trade-off baked into the architecture, meaning anyone analyzing or reproducing Llama's layers should expect this simpler normalization rather than the LayerNorm most other transformers use.