Why it matters
Optim state often dominates memory. Understanding shapes optimizer choice.
Advertisement
The architecture
AdamW: m (fp32) + v (fp32) = 8 bytes.
Lion: 1 momentum = 4 bytes.
Adafactor: row + col stats.
Advertisement
How it works end to end
Master weights fp32 + optim state: ~12 bytes/param.
Total optim mem huge for big models.
ZeRO-1: shard just optim state.