Series · Builders · Steps 2–5
Fine-Tuning & Adaptation
Adapting a model does not have to be a heavy, complex project. This series walks the adaptation ladder — prompt, retrieval, then fine-tuning — and shows why parameter-efficient methods like LoRA are useful, what sits behind settings such as rank and alpha, how to run one on modest hardware, and which common mistakes render a fine-tuning run ineffective. The focus is on understanding rather than ready-made recipes: once it is clear what each variable does, you can adapt a model deliberately and at minimal cost.
7 pieces, in order
-
Part 1: Fine-tune, RAG, or prompt: which one, and what each costs
Step 5 · 8 min -
Part 2: Why LoRA works: the intrinsic-dimensionality story
Step 5 · 4 min -
Part 3: The LoRA family: QLoRA, DoRA, and LoRA+ — which, and when?
Step 5 · 3 min -
Part 4: LoRA hyperparameters demystified: rank, alpha, and what to set
Step 5 · 4 min -
Part 5: Fine-tune your first model on free Colab: QLoRA in about 40 lines
Step 5 · 4 min -
Part 6: Common fine-tuning pitfalls and how to debug them
Step 5 · 4 min -
Part 7: How RAG actually works
Step 2 · 9 min