max_val_steps

Description

The max_val_steps parameter specifies the maximum number of validation steps to run at the end of each epoch. These validation steps evaluate the model on the validation set and compute metrics, helping track performance over time.

Increasing max_val_steps allows Kumo to evaluate the model on a larger portion of the validation set, leading to a more reliable estimation of model performance. However, this comes at the cost of increased runtime.

Reducing max_val_steps speeds up validation, enabling Kumo to spend more time processing training samples and updating the model parameters within the same total runtime. However, this may result in noisier validation metrics, which can lead to premature stopping of training. If you reduce max_val_steps, consider adjusting the early_stopping settings by increasing min_delta and patience to mitigate the risk of stopping too early.

For detailed pseudo code illustrating the training and validation process, see the max_epochs documentation.

Supported Task Types

• All

Default Values