AI Potluck
Infrastructure / Core ML frameworks & libraries

DeepSpeed

Microsoft

DeepSpeed is a deep-learning optimization library for efficient distributed training and inference of very large models. It provides system-level innovations including ZeRO memory optimization, ZeRO-Infinity, 3D parallelism, and mixture-of-experts support, and has powered training of models like BLOOM (176B) and Megatron-Turing NLG (530B). It is maintained by Microsoft and integrates with Hugging Face Transformers, PyTorch Lightning, and others.

Verified 2026-08-13 via GitHub and the LICENSE body.

Openness

5 high confidence
5.0
license
Apache-2.0(OSI)
source
public
core-gated
ungated

Apache-2.0 license body confirmed; full source public. The repository is public and unarchived, and the README describes the whole library with no paid, enterprise or hosted tier beside it, so source is public and the core ungated.

Adoption

4 high confidence
4.0

1,112,723 PyPI downloads of `deepspeed` in the trailing 30 days, which falls in the 1M-10M a month band of the software usage scale, level 4.

Capability

5 high confidence
5.0

Foundational scale-out training stack used for many of the largest published models.

Verified 2026-08-13