Why it matters
LR shapes training. Understanding shapes design.
Advertisement
The architecture
Peak LR from sweep.
Scale with batch: sqrt or linear.
Model size: larger = smaller LR.
Advertisement
How it works end to end
Common LLM peak LR: 3e-4 (Llama), 6e-4 (Chinchilla).
muP: transfer from small.
Cosine decay to 10%.