composer.models.vit_small_patch16.model

Implements ViT-S/16 as a ComposerClassifier.

Classes

ViTSmallPatch16

Implements ViT-S/16 as a ComposerClassifier.

class composer.models.vit_small_patch16.model.ViTSmallPatch16(num_classes=1000, image_size=224, channels=3, dropout=0.0, embedding_dropout=0.0)

Bases: composer.models.base.ComposerClassifier

Implements ViT-S/16 as a ComposerClassifier.

See Training data-efficient image transformers & distillation through attention (Touvron et al, 2021) for details on ViT-S/16.

Parameters

• num_classes (int, optional) – number of classes for the model. Default: 1000.

• image_size (int, optional) – input image size. If you have rectangular images, make sure your image size is the maximum of the width and height. Default: 224.

• channels (int, optional) – number of image channels. Default: 3.

• dropout (float, optional) – 0.0 - 1.0 dropout rate. Default: 0.

• embedding_dropout (float, optional) – 0.0 - 1.0 embedding dropout rate. Default: 0.