LYCEUMAI

Learn

The whole stack fits in your head. These paths take you through everything this project knows — verified code you can run, not slides.

3 paths9 modules~5 min to first gate

Six evenings through the source

  1. 1ml/tensor.mojoRow-major tensors as plain Lists of Float64; broadcast, reduce, reshape.
  2. 2ml/autograd.mojoValue nodes, the Graph arena, ~20 op builders, backward dispatch.
  3. 3ml/loss.mojoFused stable softmax+CE as one autograd op; eps clamping done consistently.
  4. 4ml/layers.mojoDense, Conv2D via im2col, MaxPool by max-index, Embedding/LayerNorm/MHA.
  5. 5ml/optim.mojoSGD → momentum → Adam → AdamW with decoupled weight decay.
  6. 6ml/models.mojoMLP, SimpleCNN, MiniGPT structs and the fit gates that certify them.

Finished the path? Train something yourself — the grokking recipe is fully specified in ml/models.mojo, and the full 30k-epoch reproduction takes about an hour.