| from transformers import PretrainedConfig | |
| from typing import List | |
| import warnings | |
| warnings.filterwarnings("ignore") | |
| class InceptionV3Config(PretrainedConfig): | |
| model_type = "inceptionv3" | |
| def __init__(self, model_name: str = "inception_v3", num_classes: int = 3, input_size: List[int] = [3, 299, 299], interpolation: str = "bicubic", mean: List[float] = [0.5, 0.5, 0.5], std: List[float] = [0.5, 0.5, 0.5], classifier: str = "fc", has_aux: bool = True, label_offset: int = 1, classes: dict = { '0': 'nsfw_gore', '1': 'nsfw_suggestive', '2': 'safe' }, output_channels: int = 2048, use_jit=False, **kwargs): | |
| self.model_name = model_name | |
| self.num_classes = num_classes | |
| self.input_size = input_size | |
| self.interpolation = interpolation | |
| self.mean = mean | |
| self.std = std | |
| self.classifier = classifier | |
| self.has_aux = has_aux | |
| self.label_offset = label_offset | |
| self.classes = classes | |
| self.output_channels = output_channels | |
| self.use_jit = use_jit | |
| super().__init__(**kwargs) | |
| """ | |
| inceptionv3_config = InceptionV3Config() | |
| inceptionv3_config.save_pretrained("inceptionv3_config") | |
| """ |