How-to guides¶

Step-by-step guides for common tasks.

• Add a model - implement a custom architecture
• Add a processor - implement a processor, including models with different training and inference behaviour
• Train a model - run single-stage end-to-end training
• Train in stages - pretrain each component separately before finetuning
• Evaluate a model - evaluate a trained model
• Run a hyperparameter sweep - run a W&B sweep of Optuna-sampled hyperparameters across platforms

For developers, there is also a guide on the process to follow to create a new release.