| from transformers import PretrainedConfig | |
| class ViTv2Config(PretrainedConfig): | |
| model_type = "vitv2" | |
| def __init__( | |
| self, | |
| img_size=224, | |
| patch_size=16, | |
| embed_dim=384, | |
| depth=12, | |
| num_heads=6, | |
| mlp_ratio=4, | |
| num_register_tokens=0, | |
| init_values=None, | |
| **ignored_kwargs, | |
| ): | |
| super().__init__(**ignored_kwargs) | |
| self.depth = depth | |
| self.img_size = img_size | |
| self.embed_dim = embed_dim | |
| self.num_heads = num_heads | |
| self.mlp_ratio = mlp_ratio | |
| self.patch_size = patch_size | |
| self.init_values = init_values | |
| self.num_register_tokens = num_register_tokens | |