DeepSpeed
Project record
Maintained by PyTorch Foundation (hosted project)68, DeepSpeed Technical Steering Committee (project committers)3
DeepSpeed is an open-source PyTorch library for distributed training and inference of large models; its features include the ZeRO memory optimizations, 3D parallelism, Ulysses sequence parallelism, and mixture-of-experts support. Microsoft contributed it to LF AI & Data as an incubation project in February 2025, and in May 2025 it became a hosted project of the PyTorch Foundation, which names Microsoft as the contributor.1967
- Repository: Repository (external site: github.com)
- Documentation: Website and documentation (external site: deepspeed.ai)
- License: LICENSE (Apache 2.0) (external site: github.com)
- Release notes: PyPI releases (external site: pypi.org)
Availability and license
Overall availability
Documented as available to the general public. Access conditions and license terms may still apply.15
Availability is separate from permission: read the license before using or redistributing.
The project charter requires inbound and outbound code under Apache 2.0 with a Developer Certificate of Origin sign-off, and makes documentation available under CC BY 4.0. The charter still names LF AI & Data as the directed fund the Technical Steering Committee communicates with, although the project is now listed by the PyTorch Foundation.38
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| Source codeIs the source code publicly readable? | Public | 1 |
| DocumentationIs user documentation published? | Public | Documentation and tutorials are on deepspeed.ai and Read the Docs.41 |
| InstallationAre installation instructions or packages publicly available? | Public | Installed with pip after PyTorch; C++/CUDA extensions are compiled just-in-time by default or can be prebuilt.14 |
| Supported platformsAre supported operating systems or hardware documented? | Public | Developed and tested mainly on NVIDIA (Pascal through Hopper) and AMD (MI100, MI200) GPUs; contributed accelerator support covers Intel Gaudi 2, Xeon CPUs, and Data Center GPU Max, Huawei Ascend NPUs, and others. Many features also run on Windows.1 |
| Release statusAre versioned releases published? | Public | Versioned releases are published on PyPI; version 0.19.7 was released on September 16, 2026.5 |
What it is useful for
Organization context
U.S. eligibility
Eligible · basis: U.S.-governed project
DeepSpeed is a PyTorch Foundation-hosted project; per the Foundation, hosted projects are governed and administered under its governance model. The PyTorch Foundation is hosted by the Linux Foundation, whose privacy policy gives a legal mailing address in San Francisco, California. The project was contributed by Microsoft, whose Form 10-K lists Redmond, Washington. The project's charter places technical oversight with a Technical Steering Committee of committers under LF Projects policies.6810311
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.