Model builder
How models work inside, from arrays to a GPT you train yourself.
At the end: Train and evaluate models, build a small GPT from scratch and fine-tune one with LoRA, knowing what every number means.
Stage 1: The maths in code
Arrays and data, the raw material.
- Python, properly: The Python every model lab uses.
- NumPy: Arrays, broadcasting and vectorised maths.
- pandas: Clean and shape the data a model learns from.
Stage 2: Learning from data
Models you can explain.
- Machine Learning: Train, evaluate and explain models honestly.
Stage 3: Inside a language model
Build it, then change it.
- Build your own GPT project: Build a small GPT with numpy only: tokenizer, attention and training.
- Fine-tuning basics: Fine-tune with LoRA on a model you can see inside.
- Evals that catch regressions: Prove the tuned model is better, not just different.