06. Module 05 Review — LLM Training Lifecycle

Focus: pretraining, SFT, RLHF / DPO, practical training knobs, and the bridge into quantization / fine-tuning efficiency.

Review loop

1. Skim the TOC in 02_explainer.md and mark the chapter that still feels least automatic.
2. Re-read only that fuzzy chapter.
3. Re-answer the self-check questions in 01_weekly_plan.md from memory.
4. Re-do the hardest prompts in 04_daily_recall.md.
5. Sketch the failure-fix table from explainer §6.1 — all 12 rows — without looking.
6. Review 05_hands_on_lab.md and write one note on data quality, one on training knobs, and one on evaluation.

Reflection

• Which stage of the lifecycle now feels operational instead of hand-wavy?
• Where are you still fuzzy: curriculum, shadowing, performance review, or specialization?
• What should feel automatic before starting 06_adaptation_compression?

Completion gate

• [ ] All explainer chapters read at least once
• [ ] Can explain the full lifecycle with the employee-training analogy
• [ ] Failure-fix table sketched from memory (all 12 rows)
• [ ] Weekly plan completed
• [ ] GPT-2 hands_on_lab shipped with before/after perplexity
• [ ] Can compute parameter memory from count × bytes
• [ ] Ready to move to 06_adaptation_compression