๐Ÿ’พ Installation#

Composer is available with pip:

pip install mosaicml

as well as with Anaconda:

conda install -c mosaicml mosaicml

To include non-core dependencies that are required by some algorithms, callbacks, datasets, or models, the following installation targets are available:

  • pip install 'mosaicml[dev]': Installs development dependencies, which are required for running tests and building documentation.

  • pip install 'mosaicml[deepspeed]': Installs Composer with support for deepspeed.

  • pip install 'mosaicml[nlp]': Installs Composer with support for NLP models and algorithms.

  • pip install 'mosaicml[wandb]': Installs Composer with support for wandb.

  • pip install 'mosaicml[comet_ml]': Installs Composer with support for comet_ml.

  • pip install 'mosaicml[neptune]': Installs Composer with support for neptune.

  • pip install 'mosaicml[tensorboard]': Installs Composer with support for tensorboard.

  • pip install 'mosaicml[streaming]': Installs Composer with support for streaming.

  • pip install 'mosaicml[mlflow]': Installs Composer with support for mlflow.

  • pip install 'mosaicml[oci]': Installs Composer with support for oci.

  • pip install 'mosaicml[onnx]': Installs Composer with support for onnx.

  • pip install 'mosaicml[coco]': Installs Composer with support for coco.

  • pip install 'mosaicml[libcloud]': Installs Composer with support for libcloud.

  • pip install 'mosaicml[all]': Installs all optional dependencies.

For a developer install, clone directly:

git clone https://github.com/mosaicml/composer.git
cd composer
pip install -e ".[all]"

Note

For fast loading of image data, we highly recommend installing Pillow-SIMD. To install, vanilla pillow must first be uninstalled.

pip uninstall pillow && pip install pillow-simd

Pillow-SIMD is not supported for Apple M1 Macs.

Docker#

To simplify environment setup for Composer, we provide a set of pre-built Docker images.

Composer Images#

The mosaicml/composer images contain Composer pre-installed with all dependencies for both NLP and Vision models. They are built on top of the mosaicml/pytorch family of images. (See the section on MosaicML PyTorch Images below.)

Note: Only the Dockerimage for most recent version of Composer will be maintained. We recommend using mosaicml/composer:latest or mosaicml/composer:latest_cpu, which will always be up to date.

Composer Version

CUDA Support

Docker Tag

0.21.1

Yes

mosaicml/composer:latest, mosaicml/composer:0.21.1

0.21.1

No

mosaicml/composer:latest_cpu, mosaicml/composer:0.21.1_cpu

Note: For a lightweight installation, we recommended using a MosaicML PyTorch Image and manually installing Composer within the image.

PyTorch Images#

The mosaicml/pytorch images contain PyTorch preinstalled, without Composer. To install composer, once inside the image, run pip install mosaicml.

Linux Distro

Flavor

PyTorch Version

CUDA Version

Python Version

Docker Tags

Ubuntu 20.04

Base

2.2.1

12.1.1 (Infiniband)

3.11

mosaicml/pytorch:2.2.1_cu121-python3.11-ubuntu20.04

Ubuntu 20.04

Base

2.2.1

12.1.1 (EFA)

3.11

mosaicml/pytorch:2.2.1_cu121-python3.11-ubuntu20.04-aws

Ubuntu 20.04

Base

2.2.1

cpu

3.11

mosaicml/pytorch:2.2.1_cpu-python3.11-ubuntu20.04

Ubuntu 20.04

Base

2.1.2

12.1.1 (Infiniband)

3.10

mosaicml/pytorch:latest, mosaicml/pytorch:2.1.2_cu121-python3.10-ubuntu20.04

Ubuntu 20.04

Base

2.1.2

12.1.1 (EFA)

3.10

mosaicml/pytorch:latest-aws, mosaicml/pytorch:2.1.2_cu121-python3.10-ubuntu20.04-aws

Ubuntu 20.04

Base

2.1.2

cpu

3.10

mosaicml/pytorch:latest_cpu, mosaicml/pytorch:2.1.2_cpu-python3.10-ubuntu20.04

Ubuntu 20.04

Base

2.0.1

11.8.0 (Infiniband)

3.10

mosaicml/pytorch:2.0.1_cu118-python3.10-ubuntu20.04

Ubuntu 20.04

Base

2.0.1

11.8.0 (EFA)

3.10

mosaicml/pytorch:2.0.1_cu118-python3.10-ubuntu20.04-aws

Ubuntu 20.04

Base

2.0.1

cpu

3.10

mosaicml/pytorch:2.0.1_cpu-python3.10-ubuntu20.04

Note: The mosaicml/pytorch:latest, mosaicml/pytorch:latest_cpu, and mosaicml/pytorch:latest-aws images will always point to the stable version of PyTorch which we have battle tested and recommend for production use. The latest label may not point to an image with the most recent release of PyTorch, however we do update our images frequently so that newer versions can be proven out.

Note: Only the images listed in the table above are maintained. All other images in the DockerHub repository have been deprecated and are kept for legacy support. Legacy images might be cleaned up without notice so itโ€™s best to migrate to the latest image or re-tag and maintain a private copy if you wish to continue using legacy images.

Pulling Images#

Pre-built images can be pulled from MosaicMLโ€™s DockerHub Repository.

For example:

docker pull mosaicml/composer