Book II · Chapter 2
Optimization and schedules
Move from SGD and Adam to schedules, clipping, and initialization.
Read first: Objectives and regularization
Chapter contents
7 entries · read in order- 01 SGD with momentum✓ Concept
- 02 Adam, AdamW, and modern optimizer choices✓ Concept
- 03 Learning rate schedules: warmup and cosine decay✓ Concept
- 04 WSD and WSD-S learning rate schedules✓ Concept
- 05 Microannealing and midtraining✓ Concept
- 06 Gradient clipping✓ Concept
- 07 Weight initialization (Kaiming, Xavier)✓ Concept