DecodeAI
Gradient Descent — Topic Deep Dive
Free

Optimizers in Practice

From SGD to AdamW with schedules that work.

Cheatsheet — Optimizers

OptimizerNotes
SGD+momentumGreat generalization often
AdamWFast start; watch generalization
WarmupStabilizes large-batch / transformers

Log LR, grad norm, and weight norm.