RAVE uses gin-config to define and override model hyperparameters without modifying code

RAVE’s hyperparameters (architecture, loss weights, training schedule, augmentations) are specified in .gin configuration files rather than command-line flags or hard-coded constants. The rave train --config v2 flag parses the corresponding gin file. Configs are composable: --config discrete --config causal applies two gin files sequentially, letting you mix architecture and mode settings. Individual bindings can be overridden at the command line with --override 'RAVE.update_discriminator_every = 4'. This makes RAVE experiments reproducible: the gin config is saved to the run directory and re-parsed at export/inference time to reconstruct the exact architecture.

Examples

rave train --config v2 --augment mute --augment compress, with individual bindings changed via --override.

Assessment

A researcher wants to change the discriminator update frequency to 4 without editing any gin file. Write the exact command-line override they should add to their rave train invocation.