Fine-Tuning in AI Engineering
Every Fine-Tuning lesson in the AI Engineering course, in the order you study them. Each page opens about the first half of its lesson to read, some the whole of it; the full lessons and their workbooks are in the plan.
Fine-tuning versus retrieval: what weight updates change, and when to use them
Fine-tuning vs RAG: what fine-tuning changes in a model, whether it can teach new facts as well as retrieval, what it costs, and when to fine-tune.
LoRA: low-rank adapters, rank and alpha, trained on a laptop
LoRA fine-tuning explained: train two thin matrices instead of the model, the code in a dozen lines, a habit taught on a laptop, and the rank and alpha knobs.
Direct preference optimization: the DPO loss against a reference model
Direct preference optimization explained: learning from a chosen and a rejected answer, the DPO loss against a frozen reference model, and preference data.
GRPO: group-relative advantages from verifiable rewards
GRPO explained: reinforcement learning from a score, sampling a group of answers per problem, group-relative advantages, and training on word problems.
Evaluating a fine-tune: target acquisition versus catastrophic forgetting
How to evaluate a fine-tuned model: did it learn the target on held-out data, and what did it lose? Catastrophic forgetting measured, with a harness.
Merge a LoRA Adapter and Export to GGUF
Page coming soon
