InceptionNeXtConfig¶
- class lucid.models.InceptionNeXtConfig(num_classes: int = 1000, depths: tuple[int, ...] | list[int] = (3, 3, 9, 3), dims: tuple[int, ...] | list[int] = (96, 192, 384, 768), token_mixers: object | None = None, mlp_ratios: int | tuple[int, ...] | list[int] = (4, 4, 4, 3), head_fn: object | None = None, drop_rate: float = 0.0, drop_path_rate: float = 0.0, ls_init_value: float = 1e-06)¶
InceptionNeXtConfig stores the stage layout and classifier settings used by
lucid.models.InceptionNeXt. It defines the per-stage depth profile,
stage widths, token mixers, MLP ratios, and regularization values.
Class Signature¶
@dataclass
class InceptionNeXtConfig:
num_classes: int = 1000
depths: tuple[int, ...] | list[int] = (3, 3, 9, 3)
dims: tuple[int, ...] | list[int] = (96, 192, 384, 768)
token_mixers: object | None = None
mlp_ratios: int | tuple[int, ...] | list[int] = (4, 4, 4, 3)
head_fn: object | None = None
drop_rate: float = 0.0
drop_path_rate: float = 0.0
ls_init_value: float = 1e-6
Parameters¶
num_classes (int): Number of output classes.
depths: Per-stage block counts for the InceptionNeXt hierarchy.
dims: Per-stage channel widths for the InceptionNeXt hierarchy.
token_mixers: Callable token mixer, or sequence of token mixer callables, used by each stage.
mlp_ratios: Positive MLP expansion ratio, or per-stage sequence of expansion ratios.
head_fn: Callable used to construct the classifier head.
drop_rate (float): Dropout probability used by the classifier head.
drop_path_rate (float): Global drop-path rate distributed across all blocks.
ls_init_value (float): Initial layer-scale value used inside each block.
Validation¶
num_classes must be greater than 0.
depths and dims must be non-empty sequences of positive integers with the same length.
token_mixers must be a callable or a same-length sequence of callables.
mlp_ratios must be a positive integer or a same-length sequence of positive integers.
head_fn must be callable.
drop_rate must be in the range [0, 1).
drop_path_rate must be in the range [0, 1].
ls_init_value must be greater than or equal to 0.
Usage¶
import lucid.models as models
config = models.InceptionNeXtConfig(
num_classes=10,
depths=(2, 2, 6, 2),
dims=(40, 80, 160, 320),
mlp_ratios=2,
drop_path_rate=0.1,
)
model = models.InceptionNeXt(config)