ƒ() Functional#

The simplest way to use Composer’s algorithms is via the functional API. Composer’s algorithms can be grouped into three, broad classes:

data augmentations add additional transforms to the training data.
model surgery algorithms modify the network architecture.
training loop modifications change the logic in the training loop.

Data augmentations can be inserted either into the dataloader as a transform or after a batch has been loaded, depending on what the augmentation acts on. Here is an example of using 🎲 RandAugment with the functional API.

import torch
from torchvision import datasets, transforms

from composer import functional as cf

c10_transforms = transforms.Compose([cf.randaugment(), # <---- Add RandAugment
                                    transforms.ToTensor(),
                                    transforms.Normalize(mean, std)])

dataset = datasets.CIFAR10('../data',
                        train=True,
                        download=True,
                        transform=c10_transforms)
dataloader = torch.utils.data.DataLoader(dataset, batch_size=1024)

Other data augmentations, such as CutMix, act on a batch of inputs. These can be inserted in the training loop after a batch is loaded from the dataloader as follows:

from composer import functional as cf

cutmix_alpha = 1
num_classes = 10
for batch_idx, (data, target) in enumerate(dataloader):
    data = cf.cutmix(  # <-- insert cutmix
        data,
        target,
        alpha=cutmix_alpha,
        num_classes=num_classes
    )
    optimizer.zero_grad()
    output = model(data)
    loss = loss_fn(output, target)
    loss.backward()
    optimizer.step()

Model surgery algorithms make direct modifications to the network itself. Functionally, these can be called as follows, using BlurPool as an example

import torchvision.models as models

from composer import functional as cf

model = models.resnet18()
cf.apply_blurpool(model)

Each method card has a section describing how to use these methods in your own trainer loop.

Functional API for applying algorithms in your own training loop.

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_squeeze_excite(model, latent_channels=64, min_channels=128)

Functions

`apply_agc`	Clips all gradients in model based on ratio of gradient norms to parameter norms.
`apply_alibi`	Removes position embeddings and replaces the attention function and attention mask as per `Alibi`.
`apply_blurpool`	Add anti-aliasing filters to the strided `torch.nn.Conv2d` and/or `torch.nn.MaxPool2d` modules within model.
`apply_channels_last`	Changes the memory format of the model to torch.channels_last.
`apply_factorization`	Replaces `Linear` and `Conv2d` modules and with `FactorizedLinear` and `FactorizedConv2d` modules.
`apply_ghost_batchnorm`	Replace batch normalization modules with ghost batch normalization modules.
`apply_squeeze_excite`	Adds Squeeze-and-Excitation blocks (Hu et al, 2019) after `Conv2d` layers.
`apply_stochastic_depth`	Applies Stochastic Depth (Huang et al, 2016) to the specified model.
`augmix_image`	Applies AugMix (Hendrycks et al, 2020) data augmentation to a single image or batch of images.
`colout_batch`	Applies ColOut augmentation to a batch of images and (optionally) targets, dropping the same random rows and columns from all images and targets in a batch.
`compute_ema`	Updates the weights of `ema_model` to be closer to the weights of `model` according to an exponential weighted average.
`cutmix_batch`	Create new samples using combinations of pairs of samples.
`cutout_batch`	See `CutOut`.
`freeze_layers`	Progressively freeze the layers of the network in-place during training, starting with the earlier layers.
`mixup_batch`	Create new samples using convex combinations of pairs of samples.
`randaugment_image`	Randomly applies a sequence of image data augmentations (Cubuk et al, 2019) to an image or batch of images.
`resize_batch`	Resize inputs and optionally outputs by cropping or interpolating.
`select_using_loss`	Prunes minibatches as a subroutine of SelectiveBackprop.
`set_batch_sequence_length`	Set the sequence length of a batch.
`should_selective_backprop`	Decides if selective backprop should be run based on time in training.
`smooth_labels`	Shrink targets towards a uniform distribution as in Szegedy et al.