EfficientNetV2Config¶

class lucid.models.EfficientNetV2Config(block_cfg: tuple[tuple[object, ...], ...] | list[tuple[object, ...]] | list[list[object]], num_classes: int = 1000, dropout: float = 0.2, drop_path_rate: float = 0.2)¶

EfficientNetV2Config stores the block sequence and classifier settings used by lucid.models.EfficientNet_V2. It defines the fused and MBConv layout, classifier size, and dropout schedule for EfficientNet-v2 variants.

Class Signature¶

@dataclass
class EfficientNetV2Config:
    block_cfg: tuple[tuple[object, ...], ...] | list[tuple[object, ...]] | list[list[object]]
    num_classes: int = 1000
    dropout: float = 0.2
    drop_path_rate: float = 0.2

Parameters¶

block_cfg: Sequence of block specs in the form (fused, out_channels, kernel_size, stride, expansion, repeats, se_scale).
num_classes (int): Number of output classes.
dropout (float): Dropout probability applied before the classifier.
drop_path_rate (float): Progressive drop-path rate applied across the repeated blocks.

Validation¶

block_cfg must contain at least one seven-value block spec.
Each block spec must use a boolean fused flag.
out_channels, kernel_size, stride, expansion, and repeats must be positive integers.
se_scale must be a non-negative integer.
num_classes must be greater than 0.
dropout must be in the range [0, 1).
drop_path_rate must be in the range [0, 1].

Usage¶

import lucid.models as models

config = models.EfficientNetV2Config(
    block_cfg=[
        (True, 24, 3, 1, 1, 1, 0),
        (False, 48, 3, 2, 4, 2, 4),
    ],
    num_classes=10,
    dropout=0.2,
    drop_path_rate=0.1,
)
model = models.EfficientNet_V2(config)