nexuml_library.scenarios.vision.cifar_resnet

nexuml_library.scenarios.vision.cifar_resnet

Composed CIFAR + ResNet classification scenarios.

cifar_resnet

cifar_resnet(
    dataset: str = "cifar10",
    download: bool = True,
    resnet_type: str = "resnet18",
    pretrained: bool = False,
    cifar_stem: bool | None = None,
    lr: float = 0.001,
    batch_size: int = 64,
    max_epochs: int = 10,
) -> ScenarioSpec

CIFAR image classification with ResNet backbone.

Parameters:

Name	Type	Description	Default
dataset	str	"cifar10" or "cifar100".	'cifar10'
download	bool	Auto-download dataset if not present.	True
resnet_type	str	ResNet variant (e.g. "resnet18").	'resnet18'
pretrained	bool	Load pretrained ImageNet weights.	False
cifar_stem	bool | None	Use CIFAR-friendly stem. Defaults to True when pretrained=False.	None
lr	float	Learning rate.	0.001
batch_size	int	Training batch size.	64
max_epochs	int	Maximum training epochs.	10

Returns:

Name	Type	Description
ScenarioSpec	ScenarioSpec	Assembled scenario with pipeline, training, data and evaluation.

Raises:

Type	Description
ValueError	If dataset is not "cifar10" or "cifar100".