coreai_opt.pruning.MagnitudePrunerConfig¶

class coreai_opt.pruning.MagnitudePrunerConfig[source]¶

Bases: CompressionConfig[ModuleMagnitudePrunerConfig]

Top-level configuration for magnitude pruning.

global_config¶

Default pruning config applied to all modules.

Type:: ModuleMagnitudePrunerConfig | None

module_type_configs¶

Per-module-type overrides.

Type:: dict[str, ModuleMagnitudePrunerConfig | None]

module_name_configs¶

Per-module-name overrides (highest priority).

Type:: dict[str, ModuleMagnitudePrunerConfig | None]

Example

>>> config = MagnitudePrunerConfig()  # 50 % sparsity everywhere
>>> config = MagnitudePrunerConfig(
...     module_name_configs={"layer1": None}  # skip layer1
... )