Container Images
CoreWeave provides containers for machine learning applications that are tuned specifically for use on the CoreWeave platform. Below is an overview on the available container images, as well as how to access them. For up-to-date version information and more details, refer to the ml-containers repository.
The ml-containers repository
The ml-containers repository contains multiple Dockerfiles, each with their own folders that contain all other files that are required for building container images using Docker.
A few special PyTorch images are available in that repository:
To view all available published images, see the ml-container repository's packages list.
PyTorch base images
ghcr.io/coreweave/ml-containers/torch
The container ml-containers/torch provided by CoreWeave contains custom builds of PyTorch, torchvision, and torchaudio, each specifically tuned for use on our platform.
There are two variants available in the ml-containers repository, respectively named base and nccl. Both base and nccl variants are available for ml-containers/torch-extras, whose versions match those provided for ml-containers/torch.
base variant
This image only includes essentials (CUDA, torch, torchvision, torchaudio). Because of this, it has a small image size, making it fast to launch.
nccl variant
This image includes development libraries and build tools, such as nvcc, which are necessary for compiling other PyTorch extensions. Because of this, it has a slightly larger image size.
nccl variant builds are built on component libraries optimized for the CoreWeave Cloud; see coreweave/nccl-tests.
PyTorch extras
ghcr.io/coreweave/ml-containers/torch-extras
The image ml-containers/torch-extras extends the ml-containers/torch images with a set of common PyTorch extensions:
Each extension is compiled against the custom PyTorch builds in ml-containers/torch.
The base edition retains a small size due to using a multi-stage build in order to avoid including CUDA development libraries in it, despite those libraries being required to build the extensions themselves.
PyTorch Nightly
ghcr.io/coreweave/ml-containers/nightly-torch
ml-containers/nightly-torch is an experimental, nightly release channel of the PyTorch Base Images in the style of PyTorch's own nightly preview builds, featuring the latest development versions of torch, torchvision, and torchaudio pulled daily from GitHub and compiled from source.
PyTorch Nightly containers are based on unstable, experimental preview builds of PyTorch, and should be expected to contain bugs and other issues. For more stable containers, use the PyTorch base and PyTorch extras containers.
PyTorch Nightly extras
ml-containers/nightly-torch-extras is a version of PyTorch Extras built on top of the ml-containers/nightly-torch container images. These are not nightly versions of the extensions themselves; they match the extension versions in the regular PyTorch Extras containers.
Additional resources
For up-to-date version information and more details on all images discussed here, please refer to the ml-containers repository.