Optimizer and learning-rate schedule study on a small model

Compare SGD with momentum and AdamW under warmup-plus-cosine and step schedules with matched tuning budgets and several seeds, logging gradient norms and update sizes.

The exercise happens on Self-directed project (Dive into Deep Learning code as a starting point). Canonical link: https://d2l.ai/chapter_optimization/index.html. Link check: link responded (5 Oct 2026). Opened from this app 0 times; opening is not counted as an attempt.

Supports: Optimization intuition backed by your own controlled evidence, for research and research-engineering interviews.