Module: tfm.core.base_trainer

Standard Trainer implementation.

The base trainer implements the Orbit StandardTrainable and StandardEvaluable interfaces. Trainers inside this project should be interchangable and independent on model architectures and tasks.

Classes

class ExperimentConfig: Top-level configuration.

class Recovery: Built-in model blowup recovery module.

class Trainer: Implements the common trainer shared for TensorFlow models.

class TrainerConfig: Configuration for trainer.

Functions

get_runtime_options(...): Get tf.distribute.RunOptions from config.