How to Train Your Model
Fine-tune your own AI on a Mac Studio or an RTX 4090 — like taming a dragon, but with data.
About This Course
Every AI you use today is someone else's dragon. This course teaches you to raise your own: pick a realistic training goal, build a clean dataset, and fine-tune an open model with LoRA/QLoRA on real consumer hardware — we follow two setups throughout, a Mac Studio M3 Ultra with 96 GB unified memory and an i9 PC with an RTX 4090. You will train "Ember", a customer-support model with your voice and your rules, evaluate it like an engineer, quantize it to GGUF, and serve it locally with Ollama or LM Studio.
What You Will Learn
- When to fine-tune vs when to use RAG or better prompting — and how to define a measurable training goal
- The exact memory math for full fine-tuning, LoRA and QLoRA — what fits on 24 GB VRAM vs 96 GB unified memory
- How to build a chat-format JSONL dataset that actually teaches behavior, and split it train/valid/test
- How to run real LoRA fine-tunes with MLX on Apple Silicon and Unsloth on NVIDIA, then evaluate and iterate
- How to level up with QLoRA and DPO, quantize to GGUF, and serve your model locally with Ollama or LM Studio
I believe AI is the most transformative technology of our time, and these tools will only get better. But being optimistic about the future doesn't mean ignoring the present — some things work beautifully, others need a human eye. This course teaches you to use AI as a power tool, not a magic wand.
Curriculum
Meet Your Dragon: Why Train Your Own Model?
Every AI model you have ever used — ChatGPT, Claude, Gemini — is someone else's dragon. Powerful, impressive, and completely under someone else's control. Th...
Set Your Training Goal: Prompt, RAG, or Fine-Tune?
The most expensive mistake in local AI is not a failed training run — it's a successful training run that solves the wrong problem. Before your dragon eats a...
Know Your Dragon's Wings: M3 Ultra 96 GB vs RTX 4090
A dragon's size is limited by its wingspan; your model's size is limited by memory. This lesson gives you the exact memory math for training, then compares o...
Dragon Food: Building a Dataset That Actually Works
You can have perfect hardware and perfect hyperparameters, and a bad dataset will still give you a bad model — every single time. Data quality is the highest...
First Flight: Train Your First LoRA
Today Ember flies. You'll pick a base model, attach LoRA adapters, and run a real fine-tune on your own machine — the Mac path with MLX, the PC path with Uns...
Taming the Beast: Evaluate, Diagnose, Iterate
Anyone can run a training script. The difference between a trained dragon and a lucky one is evaluation — measuring, diagnosing, and iterating like an engine...
Advanced Dragon Training: QLoRA, DPO, and Quantization
Ember flies and follows orders. This lesson is the advanced flight school: training bigger dragons on the same hardware (QLoRA), teaching judgment instead of...
Release Your Dragon: Deploy, Serve, and Share
Ember is trained, tested, and quantized. The final lesson is about letting it fly in the real world: serving it as an app your tools can talk to, keeping it ...