Academy / Generative AI Engineering

Path 3 · Course 3 · For developers and ML practitioners

Fine-tuning & Model Optimisation

When to fine-tune versus prompt or use RAG; full fine-tuning, LoRA and other methods; preparing training data; running and evaluating a fine-tune; and making models smaller, faster and cheaper — with Python code.

📘 6 modules⏱ About 3.5 hours✅ Knowledge check per module🎓 Final assessment & certificate🎯 Python; how LLMs work

0 of 6 modules complete

Start Module 1 Final assessment

Modules

  1. 1

    When to fine-tune

    What fine-tuning is, the prompt → model → RAG → fine-tune ladder, strengths and weaknesses, where to fine-tune, and total cost.

    25 min
  2. 2

    Fine-tuning methods

    Full fine-tuning, LoRA and QLoRA, SFT, preference tuning (DPO), reinforcement fine-tuning, continued pre-training, distillation and choosing a base model.

    30 min
  3. 3

    Preparing training data

    Why data quality dominates, the JSONL chat format, qualities of good data, sources including synthetic data, validation and splits in Python, and personal data.

    30 min
  4. 4

    Running a fine-tune

    A LoRA fine-tune in Python with open-source libraries, key hyperparameters, reading loss curves, hosted fine-tuning, and forgetting and safety erosion.

    35 min
  5. 5

    Evaluating fine-tuned models

    Baselines, task metrics, comparing models in code, regression checks for ability, safety, fairness and memorisation, blind review, A/B tests and documentation.

    30 min
  6. 6

    Making models smaller, faster and cheaper

    Quantisation, distillation, pruning and smaller models, routing, caching and batching, a cost calculation in Python and a decision checklist.

    30 min