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
- Fix ZeRO-3 checkpoint conversion with --debug
- Add Native Windows Support
- Restore memory-mapped datasets when unpickling
- Skip absent buffers when loading AutoTP checkpoints
- Accept gradient iterators in global norm helpers
- fix(zenflow): use optimizer_z3.device instead of hardcoded cuda
- Keep Random-LTD sequence lengths above the configured minimum
- Wait for all async all-gathers in coalesced ZeRO-3 parameter gather
- fix(data_pipeline): default device to active accelerator in DistributedDataAnalyzer
- Normalize Gram Newton-Schulz inputs before casting to fp16
- Docs
- Python not yet supported