Z1/2 init: flatten params on device#7828

Merged

stas00 merged 24 commits intodeepspeedai:masterfrom

ksugama:flatten-tensor-gpu

Feb 13, 2026

Commits on Feb 9, 2026

Improve engine's cleanup (#7813)

tohtana
authored and
ksugama
committed
Ignore evoformer test (#7815)

tohtana
authored and
ksugama
committed
Fix typos in accelerator setup guide (#7818)

nathon-lee
authored and
ksugama
committed
Raise clear error on in-place GatheredParameters edits without modifier_rank (#7817)

tohtana
authored and
ksugama
committed
[Bugfix] Resolve Rank index out of range during BWD when sp_size < world_size in Ulysses (#7809)

authored and
ksugama
committed
Update PyTorch to v2.9 for modal tests (#7816)

tohtana
authored and
ksugama
committed
Update version.txt to 0.18.6 after latest release (#7826)

loadams
authored and
ksugama
committed
Fix leaf module race condition (#7825)

tohtana
authored and
ksugama
committed
Skip sequence parallel operations during eval (#7821)

jp1924
authored and
ksugama
committed
Support custom partitioning patterns for AutoTP (#7806)

tohtana
authored and
ksugama
committed
flatten gpu side
ksugama
committed
repro script
ksugama
committed
detect gpu count in repro
ksugama
committed
add .venv to path
ksugama
committed
clean up
ksugama
committed
format and delete repro script
ksugama
committed
add dedicated test
ksugama
committed
parametrize tests
ksugama
committed
Fix gradient is ready with z2 (#7829)

sfc-gh-truwase
authored and
ksugama
committed
Fix AutoTP custom patterns: respect use_default_specs (#7827)

tohtana
authored and
ksugama
committed
Merge branch 'master' into flatten-tensor-gpu
ksugama
authored

Commits on Feb 10, 2026

Merge branch 'master' into flatten-tensor-gpu
ksugama
authored

Commits on Feb 12, 2026