DeepSpeed
Software
Training
Active
Microsoft's scaling toolkit — ZeRO sharding made huge models trainable.
You're looking at a software record: a tool or platform in the AI stack — who maintains it and what hardware it runs on.
DeepSpeed's ZeRO optimizer sharding let models far larger than one GPU's memory train efficiently, powering the Megatron-Turing era and countless open efforts since. Its ideas (now echoed by PyTorch FSDP) defined how the industry splits models across thousands of devices.
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Maintained by
Related
Sources
Record reviewed as of 2026-01-15 — see Methodology.