composer.functional

Algorithms can be used directly through our functions-based API.

from composer import functional as CF
from torchvision import models

model = models.resnet50()

# replace some layers with blurpool
CF.apply_blurpool(model)
# replace some layers with squeeze-excite
CF.apply_se(model, latent_channels=64, min_channels=128)

functional.augment_and_mix	Applies AugMix (Hendrycks et al.) data augmentation to an image.
functional.apply_blurpool	Add anti-aliasing filters to the strided torch.nn.Conv2d and/or torch.nn.MaxPool2d modules within model.
functional.apply_alibi	Removes position embeddings and replaces the attention function and attention mask according to AliBi.
functional.colout	Drops random rows and columns from a single image.
functional.cutout	See CutOut.
functional.smooth_labels	Shrinks targets towards a uniform distribution to counteract label noise as in Szegedy et al..
functional.freeze_layers	Progressively freeze the layers of the network during training, starting with the earlier layers.
functional.apply_ghost_batchnorm	Replace batch normalization modules with ghost batch normalization modules.
functional.mixup_batch	Create new samples using convex combinations of pairs of samples.
functional.resize_inputs	Resize inputs and optionally outputs by cropping or interpolating.
functional.randaugment	Randomly applies a sequence of image data augmentations (Cubuk et al. 2019).
functional.scale_scheduler	Makes a learning rate schedule take a different number of epochs.
functional.selective_backprop	Select a subset of the batch on which to learn as per (Jiang et al. 2019).
functional.apply_se	See SqueezeExcite
functional.apply_seq_length_warmup	Progressively increases the sequence length during training.
functional.apply_stochastic_depth	Applies Stochastic Depth (Huang et al.) to the specified model.