deepspeed
https://github.com/microsoft/deepspeed
Python
DeepSpeed is a deep learning optimization library that makes distributed training easy, efficient, and effective.
Triage Issues!
When you volunteer to triage issues, you'll receive an email each day with a link to an open issue that needs help in this project. You'll also receive instructions on how to triage issues.
Triage Docs!
Receive a documented method or class from your favorite GitHub repos in your inbox every day. If you're really pro, receive undocumented methods or classes and supercharge your commit history.
Python not yet supported21 Subscribers
Add a CodeTriage badge to deepspeed
Help out
- Issues
- Clear the gradient ZenFlow just offloaded, not param.grad
- feat(rollout): HybridEngineRollout graph capture + segment-KI for Qwen3.5 hybrid — 67 tok/s (1.14× of vLLM)
- fix: skip None buffers in AutoTP Loading.load_buffer (checkpoint load with nn.InstanceNorm1d)
- [REQUEST] NPU: implement graph operations for short kernel sequences (follow-up to #4318)
- Implement graph operations for NPU accelerator
- [BUG] Skip None buffers in AutoTP checkpoint loading
- Follow the optimizer's own state names in universal checkpoints
- Nightly full-suite run with classified failure triage and automated git-bisect
- Universal checkpoint conversion only knows Adam's state keys, so a Muon checkpoint cannot be converted
- [BUG]_prepare_fp32_grad_for_sub_group clears every parameter's gradient once per sub-group
- Docs
- Python not yet supported