Expand description
Optimizers.
This module provides small, allocation-free-per-step optimizers that update an Mlp
given a set of Gradients.
Design notes:
- Optimizer state (momentum/Adam moments) lives outside the model.
- The training loop owns the optimizer state and reuses it across steps.
Structs§
- Sgd
- Stochastic gradient descent with a fixed learning rate.
Enums§
- Optimizer
- Optimizer choice for training.
- Optimizer
State - Owned optimizer state.