transformer_lens.model_bridge.component_setup module¶
- transformer_lens.model_bridge.component_setup.replace_remote_component(replacement_component: Module, remote_path: str, remote_model: Module) None¶
Replace a component in a remote model.
- Parameters:
replacement_component – The new component to install
remote_path – Path to the component in the remote model
remote_model – The remote model to modify
- transformer_lens.model_bridge.component_setup.set_original_components(bridge_module: Module, architecture_adapter: ArchitectureAdapter, original_model: Module) None¶
Set original components on the pre-created bridge components.
- Parameters:
bridge_module – The bridge module to configure
architecture_adapter – The architecture adapter
original_model – The original model to get components from
- transformer_lens.model_bridge.component_setup.setup_blocks_bridge(blocks_template: Any, architecture_adapter: ArchitectureAdapter, original_model: Module) ModuleList¶
Set up blocks bridge with proper ModuleList structure.
- Parameters:
blocks_template – Template bridge component for blocks
architecture_adapter – The architecture adapter
original_model – The original model to get components from
- Returns:
ModuleList of bridged block components
- transformer_lens.model_bridge.component_setup.setup_components(components: dict[str, Any], bridge_module: Module, architecture_adapter: ArchitectureAdapter, original_model: Module) None¶
Set up components on the bridge module.
- Parameters:
components – Dictionary of component name to bridge component mappings
bridge_module – The bridge module to configure
architecture_adapter – The architecture adapter
original_model – The original model to get components from
- transformer_lens.model_bridge.component_setup.setup_submodules(component: GeneralizedComponent, architecture_adapter: ArchitectureAdapter, original_model: Module) None¶
Set up submodules for a bridge component using proper component setup.
- Parameters:
component – The bridge component to set up submodules for
architecture_adapter – The architecture adapter
original_model – The original model to get components from