Composer is a PyTorch library that enables you to train neural networks faster, at lower cost, and to higher accuracy. We've implemented more than two dozen speedup methods that can be applied to your training loop in just a few lines of code, or used with our built-in Trainer. We continually integrate the latest state-of-the-art in efficient neural network training.
Composer features:
- 20+ methods for speeding up training networks for computer vision and natural language. Don't waste hours trying to reproduce research papers when Composer has done the work for you.
- An easy-to-use trainer that has been written to be as performant as possible and integrates best practices for efficient, multi-GPU training.
- Functional forms of all of our speedup methods that allow you to integrate them into your existing training loop.
- Strong, reproducible baselines to get you started as quickly as possible.
With no additional tuning, you can apply our methods to:
- Train ResNet-50 on ImageNet to the standard 76.6% top-one accuracy for $15 in 27 minutes (with vanilla PyTorch: $116 in 3.5 hours) on AWS.
- Train GPT-2 125M to the standard perplexity of 24.11 for $145 in 4.5 hours (with vanilla PyTorch: $255 in 7.8 hours) on AWS.
- Train DeepLab-v3 on ADE20k to the standard mean IOU of 45.7 for $36 in 1.1 hours (with vanilla PyTorch: $110 in 3.5 hours) on AWS.
Composer is available with Pip:
pip install mosaicmlAlternatively, install Composer with Conda:
conda install -c mosaicml mosaicmlYou can use Composer's speedup methods in two ways:
- Through a standalone Functional API (similar to
torch.nn.functional) that allows you to integrate them into your existing training code. - Using Composer's built-in Trainer, which is designed to be performant and automatically takes care of the details of using speedup methods.
Integrate our speedup methods into your training loop with just a few lines of code, and see the results. Here we easily apply BlurPool and SqueezeExcite:
import composer.functional as cf
from torchvision import models
my_model = models.resnet18()
# add blurpool and squeeze excite layers
cf.apply_blurpool(my_model)
cf.apply_squeeze_excite(my_model)
# your own training code starts hereFor more examples, see the Composer Functional API Colab notebook and Functional API guide.
For the best experience and the most efficient possible training, we recommend using Composer's built-in trainer, which automatically takes care of the details of using speedup methods and provides useful abstractions that facilitate rapid experimentation.
import torch
# adaptive_avg_pool2d_backward_cuda in mnist_classifier is not deterministic
torch.use_deterministic_algorithms(False)-->
from torch.utils.data import DataLoader
from torchvision import datasets, transforms
from composer import Trainer
from composer.algorithms import ChannelsLast, CutMix, LabelSmoothing
from composer.models import mnist_model
transform = transforms.Compose([transforms.ToTensor()])
train_dataset = datasets.MNIST("data", download=True, train=True, transform=transform)
eval_dataset = datasets.MNIST("data", download=True, train=False, transform=transform)
train_dataloader = DataLoader(train_dataset, batch_size=128)
eval_dataloader = DataLoader(eval_dataset, batch_size=128)
trainer = Trainer(
model=mnist_model(),
train_dataloader=train_dataloader,
eval_dataloader=eval_dataloader,
max_duration="1ep",
algorithms=[
ChannelsLast(),
CutMix(alpha=1.0),
LabelSmoothing(smoothing=0.1),
]
)
trainer.fit()Composer's built-in trainer makes it easy to add multiple speedup methods in a single line of code! Trying out new methods or combinations of methods is as easy as changing a single list.
Here are some examples of methods available in Composer (see here for the full list):
| Name | Attribution | tl;dr | Example Benchmark | Speed Up* |
|---|---|---|---|---|
| Alibi | Press et al, 2021 | Replace attention with AliBi. | GPT-2 | 1.5x |
| BlurPool | Zhang, 2019 | Applies an anti-aliasing filter before every downsampling operation. | ResNet-101 | 1.2x |
| ChannelsLast | PyTorch | Uses channels last memory format (NHWC). | ResNet-101 | 1.5x |
| CutOut | DeVries et al, 2017 | Randomly erases rectangular blocks from the image. | ResNet-101 | 1.2x |
| LabelSmoothing | Szegedy et al, 2015 | Smooths the labels with a uniform prior | ResNet-101 | 1.5x |
| MixUp | Zhang et al, 2017 | Blends pairs of examples and labels. | ResNet-101 | 1.5x |
| RandAugment | Cubuk et al, 2020 | Applies a series of random augmentations to each image. | ResNet-101 | 1.3x |
| SAM | Foret et al, 2021 | An optimization strategy that seeks flatter minima. | ResNet-101 | 1.4x |
| SeqLengthWarmup | Li et al, 2021 | Progressively increase sequence length. | GPT-2 | 1.2x |
| Stochastic Depth | Huang et al, 2016 | Replaces a specified layer with a stochastic version that randomly drops the layer or samples during training | ResNet-101 | 1.1x |
* = time-to-train to the same quality as the baseline.
Given two methods that speed up training by 1.5x each, do they combine to provide a 2.25x (1.5x * 1.5x) speedup? Not necessarily. They may optimize the same part of the training process and lead to diminishing returns, or they may even interact in ways that prove detrimental. Determining which methods to compose together isn't as simple as assembling a set of methods that perform best individually.
We have come up with compositions of methods that work especially well together through rigorous exploration of the design space of recipes and research on the science behind composition.
As an example, here are two performant recipes, one for ResNet-101 on ImageNet, and the other for GPT-2 on OpenWebText, on 8xA100s:
| Name | Functional | tl;dr | Benchmark | Speed Up |
|---|---|---|---|---|
| Blur Pool | cf.apply_blurpool |
Applies an anti-aliasing filter before every downsampling operation. | ResNet-101 | 1.2x |
| Channels Last | cf.apply_channels_last |
Uses channels last memory format (NHWC). | ResNet-101 | 1.5x |
| Label Smoothing | cf.smooth_labels |
Smooths the labels with a uniform prior. | ResNet-101 | 1.5x |
| MixUp | CF.mixup_batch |
Blends pairs of examples and labels. | ResNet-101 | 1.5x |
| Progressive Resizing | cf.resize_batch |
Increases the input image size during training. | ResNet-101 | 1.3x |
| SAM | N/A |
SAM optimizer measures sharpness of optimization space. | ResNet-101 | 1.5x |
| Composition | N/A |
Cheapest: $49 @ 78.1% Acc | ResNet-101 | 3.5x |
| Name | Functional | tl;dr | Benchmark | Speed Up |
|---|---|---|---|---|
| Alibi | cf.apply_alibi |
Replace attention with AliBi. | GPT-2 | 1.6x |
| Seq Length Warmup | cf.set_batch_sequence_length |
Progressively increase sequence length. | GPT-2 | 1.5x |
| Composition | N/A |
Cheapest: $145 @ 24.11 PPL | GPT-2 | 1.7x |
We'll use the word benchmark to denote a specific model trained on a specific dataset, with model quality assessed using a specific metric.
Composer features computer vision and natural language processing benchmarks including (but not limited to):
| Model | Dataset | Loss | Task | Evaluation Metrics |
|---|---|---|---|---|
| Computer Vision | ||||
| ResNet Family | CIFAR-10 | Cross Entropy | Image Classification | Classification Accuracy |
| ResNet Family | ImageNet | Cross Entropy | Image Classification | Classification Accuracy |
| EfficientNet Family | ImageNet | Cross Entropy | Image Classification | Classification Accuracy |
| UNet | BraTS | Dice Loss | Image Segmentation | Dice Coefficient |
| DeepLab v3 | ADE20K | Cross Entropy | Image Segmentation | mIoU |
| Natural Language Processing | ||||
| BERT Family | {Wikipedia & BooksCorpus, C4} | Cross Entropy | Masked Language Modeling | GLUE |
| GPT Family | {OpenWebText, C4} | Cross Entropy | Language Modeling |
Perplexity |
The compute required to train a state-of-the-art machine learning model is doubling every 6 months, putting such models further and further out of reach for most researchers and practitioners with each passing day.
Composer addresses this challenge by focusing on training efficiency: it contains cutting-edge speedup methods that modify the training algorithm to reduce the time and cost necessary to train deep learning models. When you use Composer, you can rest assured that you are training efficiently. We have combed the literature, done the science, and built industrial-grade implementations to ensure this is the case.
Even after these speedup methods are implemented, assembling them together into recipes is nontrivial. We designed Composer with the right abstractions for composing (and creating new) speedup methods.
Specifically, Composer uses two-way callbacks (Howard et al, 2020) to modify the entire training state at particular events in the training loop to effect speedups. We handle collisions between methods, proper method ordering, and more.
Through this, methods can modify:
- data inputs for batches (data augmentations, sequence length warmup, skipping examples, etc.)
- neural network architecture (pruning, model surgery, etc.)
- loss function (label smoothing, MixUp, CutMix, etc.)
- optimizer (Sharpness Aware Minimization)
- training dynamics (layer freezing, selective backprop, etc.)
You can easily add your own methods or callbacks to try out your ideas or modify any part of the training loop.
Composer is an active and ongoing project. We will respond quickly to issues filed in this repository.
- Composer is mostly optimized for computer vision and natural language processing. If you work on, e.g., reinforcement learning, you might encounter rough edges when using Composer.
- Composer currently only supports NVIDIA GPUs, although we're working on adding alternatives.
- Since Composer is still in alpha, our API may not be stable. We recommend pegging your work to a Composer version.
Here are some resources actively maintained by the Composer community to help you get started:
| Resource | Details |
|---|---|
| Getting started with our Trainer | A Colab Notebook showing how to use our Trainer |
| Getting started with our Functional API | A Colab Notebook showing how to use our Functional API |
| Building Speedup Methods | A Colab Notebook showing how to build new training modifications on top of Composer |
| Training BERTs with Composer and 🤗 | A Colab Notebook showing how to train BERT models with Composer and 🤗! |
If you have any questions, please feel free to reach out to us on Twitter, email, or our Community Slack!
Composer is part of the broader Machine Learning community, and we welcome any contributions, pull requests, or issues!
To start contributing, see our Contributing page.
P.S.: We're hiring!
@misc{mosaicml2022composer,
author = {The Mosaic ML Team},
title = {composer},
year = {2021},
howpublished = {\url{https://github.com/mosaicml/composer/}},
}