OptimConfig

Contents

OptimConfig#

class stable_ssl.config.OptimConfig(optimizer: dict, scheduler: dict, epochs: int = 1000, max_steps: int = -1, accumulation_steps: int = 1, grad_max_norm: float | None = None)[source]#

Bases: object

Configuration for the optimization parameters.

Parameters:

  • optimizer (dict) – Configuration for the optimizer.

  • scheduler (dict) – Configuration for the learning rate scheduler.

  • epochs (int, optional) – Number of epochs to train the model. Default is 1000.

  • max_steps (int, optional) – Maximum number of steps to train the model. Default is -1. If negative, the models trains on the full dataset. If it is between 0 and 1, it represents the fraction of the dataset to train on.

  • accumulation_steps (int, optional) – Number of steps to accumulate gradients before updating the model. Default is 1.

  • grad_max_norm (float, optional) – Maximum norm of the gradients. If None, no clipping is applied. Default is None.

Contents