nat.profiler.parameter_optimization.optimizable_utils#

Attributes#

logger

Functions#

walk_optimizables(→ dict[str, ...)

Recursively build {flattened.path: SearchSpace} for every optimizable

Module Contents#

logger#
walk_optimizables(
obj: pydantic.BaseModel,
path: str = '',
) → dict[str, nat.data_models.optimizable.SearchSpace]#

Recursively build {flattened.path: SearchSpace} for every optimizable field inside obj.

  • Honors optimizable_params on any model that mixes in OptimizableMixin – only listed fields are kept.

  • If a model contains optimizable fields but omits optimizable_params, we emit a warning and skip them.