Training and Fine-tuning

Shape a base model: datasets, loss curves, SFT and LoRA/QLoRA, DPO, and the judgment of when to fine-tune vs prompt vs RAG.

1. 1Datasets: the raw material a model learns from3 q
2. 2Reading loss curves: overfitting and validation3 q
3. 3Supervised fine-tuning (SFT)3 q
4. 4LoRA from scratch: freeze the base, learn a little3 q
5. 5LoRA and QLoRA in practice: base vs adapter3 q
6. 6Preference tuning with DPO3 q
7. 7Evaluating a fine-tune without fooling yourself3 q
8. 8Putting it together: fine-tune vs prompt vs RAG3 q