get_adapter_state_dict

torchtune.modules.peft.get_adapter_state_dict(state_dict: Dict[str, Any], device: Optional[str] = 'cpu') → Dict[str, Any]

Return the subset of the full state_dict from a model that correspond to an adapter. Assumes that “lora” and “magnitude” are unique names for adapter parameters, and that the state_dict is not sharded. All returned parameters are moved to CPU.

Parameters:

• state_dict (Dict[str, Any]) – Full model state dict.

• device (Optional[str]) – device to move adapter parameters to. Default: ‘cpu’

Returns:

the subset of model’s state dict containing only adapter parameters.

Return type:

Dict[str, Any]