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
- Avoid optimizer updates during low-precision initialization
- Keep empty gradient-norm ranks in collective reductions
- Track implicit package initializer dependencies in CI test selection
- Fix inference normalization row bounds and small-channel dispatch
- Stop the comms logger exiting on all_gather_object and monitored_barrier
- [Proposal] Extend affine live-state transfer to CPU-offloaded optimizer storage
- Recover legacy Torch checkpoints in zero_to_fp32
- Cover ZeRO-3 DP-resize rejection before serial checkpoint loading
- Complete ZeRO-3 serial checkpoint handoff for universal loads and rank boundaries
- [BUG] Universal checkpoint resume puts CPU `step` into fused/capturable optimizer state
- Docs
- Python not yet supported