Pull Requests microsoft/DeepSpeed

Fix loss scaling and backward call of ZenFlow

#7793 by tohtana was closed 2026-01-20 17:58

Fix/dc zero3 allgather uneven shards

#7791 by juyterman1000 was closed 2026-01-18 05:05

Fix BF16_Optimizer being used without ZeRO

#7790 by tohtana was merged 2026-01-20 23:04

Skip empty parameters in gradient reduction

#7789 by tohtana was merged 2026-01-20 21:50

Fix issue with BF16 optimizer selection

#7788 by tohtana was merged 2026-01-20 22:28

Fix backward for pipeline engine

#7787 by tohtana was merged 2026-01-20 14:10

fix checkpointing/loading of z0+bf16

#7786 by tohtana was merged 2026-01-19 21:07

Merge Ulysses test fix to PR for full CI test

#7785 by tohtana was merged 2026-01-17 05:11

Fix Ulysses PEFT test

#7784 by tohtana was merged 2026-01-17 18:31

Add workflow to run full tests

#7783 by tohtana was closed 2026-01-18 23:20

Add workflow to run full tests

#7782 by tohtana was closed 2026-01-17 00:55

docs: update README.md

#7781 by eltociear was merged 2026-01-17 03:59

Fix gradient checkpointing with use_reentrant=True / PyTorch-style backward / ZeRO-3

#7780 by tohtana was merged 2026-01-17 04:55

[Docs] Fix `docs/_pages/config-json.md` format

#7779 by ooooo-create was merged 2026-01-15 07:26

Update CLA to refer to DCO

#7778 by loadams was merged 2026-01-15 07:28

Remove cron/PR triggers for outdated V100 tests

#7777 by loadams was merged 2026-01-13 15:53

Add timeout to test workflows

#7774 by tohtana was merged 2026-01-13 04:06

fix: skip compressed allreduce for empty tensors

#7769 by T1mn was merged 2026-01-15 22:18

Udpate workflow trigger

#7768 by tohtana was merged 2026-01-10 03:49

Various fixes to run on mps

#7767 by jeffra was merged 2026-01-09 23:39

fix: delete using namespace std.

#7766 by nathon-lee was merged 2026-01-10 04:27

Update version.txt after 0.18.4 release

#7765 by loadams was merged 2026-01-07 23:40

Introduce all_reduce_hook to support gradient aggregation across replica groups.

#7764 by zhengchenyu was closed 2026-03-01 20:38

fix: avoid IndexError in BF16_Optimizer.destroy() when using DummyOptim

#7763 by nathon-lee was merged 2026-01-07 08:32

fix: update Megatron-DeepSpeed tutorial to match current repo structure

#7761 by nathon-lee was merged 2026-01-11 06:48

Fix Evoformer compilation

#7760 by sdvillal was merged 2026-01-18 08:33

Fix DeepCompile for PyTorch 2.8/2.9 compatibility

#7755 by tohtana was merged 2026-01-02 23:32

Add fallback to BF16 support check

#7754 by tohtana was merged 2026-01-02 22:29

Add CI workflow to run tests on AWS

#7753 by tohtana was merged 2026-01-02 22:06

fix: remove premature MPI environment variable check in OpenMPIRunner

#7751 by nathon-lee was merged 2026-01-01 08:33