Class ModelManager

diffsynth/models/model_manager.py:316–466

class ModelManager:
    def __init__(
        self,
        torch_dtype=torch.float16,
        device="cuda",
        model_id_list: List[Preset_model_id] = [],
        downloading_priority: List[Preset_model_website] = ["ModelScope", "HuggingFace"],
        file_path_list: List[str] = [],
    ):
        self.torch_dtype = torch_dtype
        self.device = device
        self.model = []
        self.model_path = []
        self.model_name = []
        downloaded_files = download_models(model_id_list, downloading_priority) if len(model_id_list) > 0 else []
        self.model_detector = [
            ModelDetectorFromSingleFile(model_loader_configs),
            ModelDetectorFromSplitedSingleFile(model_loader_configs),
            ModelDetectorFromHuggingfaceFolder(huggingface_model_loader_configs),
            ModelDetectorFromPatchedSingleFile(patch_model_loader_configs),
        ]
        self.load_models(downloaded_files + file_path_list)


    def load_model_from_single_file(self, file_path="", state_dict={}, model_names=[], model_classes=[], model_resource=None):
        print(f"Loading models from file: {file_path}")
        if len(state_dict) == 0:
            state_dict = load_state_dict(file_path)
        model_names, models = load_model_from_single_file(state_dict, model_names, model_classes, model_resource, self.torch_dtype, self.device)
        for model_name, model in zip(model_names, models):
            self.model.append(model)
            self.model_path.append(file_path)
            self.model_name.append(model_name)
        print(f"    The following models are loaded: {model_names}.")


    def load_model_from_huggingface_folder(self, file_path="", model_names=[], model_classes=[]):
        print(f"Loading models from folder: {file_path}")
        model_names, models = load_model_from_huggingface_folder(file_path, model_names, model_classes, self.torch_dtype, self.device)
        for model_name, model in zip(model_names, models):
            self.model.append(model)
            self.model_path.append(file_path)
            self.model_name.append(model_name)
        print(f"    The following models are loaded: {model_names}.")


    def load_patch_model_from_single_file(self, file_path="", state_dict={}, model_names=[], model_classes=[], extra_kwargs={}):
        print(f"Loading patch models from file: {file_path}")
        model_names, models = load_patch_model_from_single_file(
            state_dict, model_names, model_classes, extra_kwargs, self, self.torch_dtype, self.device)
        for model_name, model in zip(model_names, models):
            self.model.append(model)
            self.model_path.append(file_path)
            self.model_name.append(model_name)
        print(f"    The following patched models are loaded: {model_names}.")


    def load_lora(self, file_path="", state_dict={}, lora_alpha=1.0):