peal.training.interfaces¶
Pydantic configs of a predictor run: how it is trained and what it is.
TrainingConfig carries the optimisation hyper-parameters and the
augmentation / robustness switches ModelTrainer reads, plus the running
counters (epoch, global_train_step) that the trainer updates in place
so a checkpoint can be resumed. PredictorConfig is the top-level schema of
a predictor.yaml: training, task, architecture and data configs, the run
directory and the checkpoint / distillation switches.
Classes
The config template for a model. |
|
Hyper-parameters and switches of one predictor training run. |