SGD Optimizer Template
sgd.h implements the minimal optimizer stage:
raw_update = -learning_rate * gradient
It does not update parameters directly.
Why Raw Updatesβ
Global throttling needs to inspect or bound the update vector before it is applied. For that reason, the optimizer emits raw update proposals:
weight_update,bias_update.
The controller/update stage later applies:
parameter += alpha * raw_update
For CTRL-NONE, alpha=1, so this behaves like plain SGD.
Config Contractβ
The generated config struct must provide:
n_in,n_out,weight_grad_t,bias_grad_t,raw_update_t,learning_rate_t.
The learning rate may eventually be either a compile-time constant or a runtime input. The current template accepts it as an argument so the top-level writer can support both policies.