Training and fine-tuning
From pretraining through post-training to the adaptation you’d actually do.
| # | File | Covers |
|---|---|---|
| 01 | Pretraining | next-token objective, data pipeline, scaling laws, base vs instruct |
| 02 | Supervised fine-tuning and instruction tuning | SFT, data quality over quantity, LoRA defaults, synthetic data |
| 03 | RLHF and preference optimization | PPO → DPO → GRPO → RLVR, and why the field moved |
| 04 | Quantization | INT4/FP8, GPTQ/AWQ, QLoRA, quality trade-offs |
| 05 | Distillation and synthetic data | teacher to student, generating and filtering data, model collapse |
| 07 | Fine-Tuning vs RAG vs Prompt Engineering | the decision most teams get wrong |
The three that get asked
Compute-optimal is not inference-optimal. Chinchilla optimises training cost. If you serve a model heavily, over-train a smaller one — you buy cheaper inference forever.
GRPO dropped the critic. PPO needs a value network to compute advantages; GRPO samples a group of responses to the same prompt and uses the group mean as the baseline. Half the memory, still on-policy. That’s why reasoning training converged on it.
Fine-tuning teaches form, not facts. Knowledge in weights can’t be updated or cited. If the ask is “answer from our documents”, it’s retrieval.