Learning library

Every guide.
One address.

Open a finished field guide or compare every on-policy distillation experiment in one place.

Field guides

  1. 01Attention & TransformersFollow one token through queries, keys, values, and multi-head attention.
  2. 02BackpropagationSee how one mistake travels backward and changes a neural network.
  3. 03EmbeddingsBuild an intuition for vector meaning, distance, and learned geometry.
  4. 04GNNs on Power GridsLearn message passing through a tiny, inspectable electrical grid.
  5. 05GraphMAEUnderstand graph representation learning by reconstructing what was hidden.
  6. 06Loss FunctionsTurn prediction errors into objectives a model can optimize.
  7. 07LSTMTrace memory, gates, and gradients through a recurrent sequence model.
  8. 08Model EvaluationAsk whether a model is actually better—and what the evidence can support.
  9. 09RolloutsConnect policy samples, trajectories, rewards, and learning signals.
  10. 10Self-supervised LearningLearn how structure in the data can provide its own supervision.
  11. 11KL DivergenceBuild KL from two forecasts, then explore extra surprise on a 3D landscape.

On-policy distillation experiments

Five independent teaching directions, each paired with its second revision. Open versions side by side to compare the iteration.