Trainer

interface Trainer

A Trainer defines a workflow for training a model.

All trainers are run incrementally, meaning that they can be run multiple times.

For each epoch, the trainer has an opportunity to review the training data with or without the student model to select a subset of data to train on.

Epoch - A single "step" in a trainer's algorithm. Some optimizers make multiple passes through datasets.

Inheritors

Properties

Link copied to clipboard
abstract val snapshot: StateFlow<Snapshot>

Exposes the latest state of training.

Functions

Link copied to clipboard
abstract fun train(iterations: Int = 1): Int

Train the model for a number of iterations, a.k.a. epochs.