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
- Measure first-call and post-update costs of continuous rollout CUDA Graphs
- Scale ZeRO-3 __avg_scatter_grads by sequence_parallel_size
- [REQUEST] Add an OPSD training batch builder for teacher-prefix assembly and response-logit alignment
- Fused CPU Adam step
- Keep the parameters a model names for fp32 in fp32 under ZeRO-3
- Universal Checkpointing: read AutoEP ZeRO-3 optimizer files one at a time
- AutoTP: send CPU copies in the first-batch consistency check
- AutoEP: support combine_impl="fused_weighted_sum" with AutoEP+AutoTP folding
- AutoEP + AutoTP folding: Universal Checkpointing under ZeRO-3
- [BUG] inference v2 layer_norm / rms_norm kernels read and write past the last row for hidden sizes <= 128 (reachable via CUDARMSNorm, e.g. EXAONE4 Q/K norm)
- Docs
- Python not yet supported