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
- Add remaining parallelism rank and world-size engine APIs
- Fix zero_to_fp32 --safe_serialization on models with tied weights
- Return frozen parameters as fp32 from get_fp32_state_dict_from_zero_checkpoint
- [BUG] set_norm_for_param_grad_in_gpu's grad_accum fallback is dead in both accumulation modes
- Validate positive inference min output tokens
- Add coalesced full gradient access
- perf(rollout): combine shared prefill with CUDA Graph decode
- ZeRO-1/2 average_tensor serializes backward compute behind the preceding reduction
- [Feature] Per-Head Muon: per-head Newton-Schulz orthogonalization for attention projections
- Reuse existing hpZeRO group in nested zero.Init contexts
- Docs
- Python not yet supported