Skip to content

Training & Model Quality

These guides cover customizations that help you iterate on model quality beyond the defaults — swapping metrics, loss functions, loading overrides, and multi-GPU training.

In This Section

Guide Description
Custom Metrics Add or replace evaluation metrics beyond the task defaults
Custom Loss & Callbacks Swap the loss function, attach callbacks, or configure early stopping
Loading Overrides Checkpoint contents, split overrides, and prediction filtering
Distributed Training Multi-GPU training with DDP and scaling tips