PR #7828 Z1/2 init: flatten params on device

Improve engine's cleanup (#7813)

ksugama committed 185 days ago

Ignore evoformer test (#7815)

ksugama committed 185 days ago

Fix typos in accelerator setup guide (#7818)

ksugama committed 185 days ago

Raise clear error on in-place GatheredParameters edits without modifier_rank (#7817)

ksugama committed 185 days ago

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

ksugama committed 185 days ago

Update PyTorch to v2.9 for modal tests (#7816)

ksugama committed 185 days ago

Update version.txt to 0.18.6 after latest release (#7826)

ksugama committed 185 days ago

Fix leaf module race condition (#7825)

ksugama committed 185 days ago

Skip sequence parallel operations during eval (#7821)

ksugama committed 185 days ago

Support custom partitioning patterns for AutoTP (#7806)

ksugama committed 185 days ago

flatten gpu side

ksugama committed 185 days ago

repro script

ksugama committed 185 days ago

detect gpu count in repro

ksugama committed 185 days ago

add .venv to path

ksugama committed 185 days ago

clean up

ksugama committed 185 days ago

format and delete repro script

ksugama committed 185 days ago

add dedicated test

ksugama committed 185 days ago

parametrize tests

ksugama committed 185 days ago

Fix gradient is ready with z2 (#7829)

ksugama committed 185 days ago

Fix AutoTP custom patterns: respect use_default_specs (#7827)

ksugama committed 185 days ago

Merge branch 'master' into flatten-tensor-gpu

ksugama committed 185 days ago

Merge branch 'master' into flatten-tensor-gpu

ksugama committed 184 days ago

Merge branch 'master' into flatten-tensor-gpu

ksugama committed 182 days ago

Merge branch 'master' into flatten-tensor-gpu

ksugama committed 182 days ago

DeepSpeed Z1/2 init: flatten params on device #7828 Merged

DeepSpeed
Z1/2 init: flatten params on device
#7828

Merged