Build a transformer

Work through attention and the components behind modern language models.

Estimated time: 180–240 minutes. Prerequisites: NumPy, matrix multiplication, and neural networks.

1. Step 1: Scaled Dot-Product Attention →
2. Step 2: Multi-Head Attention →
3. Step 3: Sinusoidal Positional Encoding →
4. Step 4: Layer Normalization →
5. Step 5: Causal Attention Mask →