microsoft/DeepSpeed

Pull Requests Commits

Merge branch 'master' of github.com:microsoft/DeepSpeed into olruwase/adam_types

tjruwase committed 4 years ago

3ca63714

Support optimizer AdamW type

tjruwase committed 4 years ago

e51e4d79

Validate consistent ckpt tags across ranks (#667)

jeffra committed 4 years ago

Verified f032e56f

Move workspace memory-allocation to PyTorch (#661)

RezaYazdaniAminabadi committed 4 years ago

Verified 981bc7d4

squash latest flops profiling changes (#1) (#664)

cli99 committed 4 years ago

Verified e2fbe4d2

Handle actvitation checkpointing args that are None or non-tensors (#660)

Shaden Smith committed 4 years ago

Verified adcfd269

LR scheduler unit tests (#429)

tjruwase committed 4 years ago

Verified da5563a9

version bump to 0.3.10

jeffra committed 4 years ago

c14b839d

Remove a very verbose print statement. (#649)

awan-10 committed 4 years ago

Verified af212f66

add additional validation checks in elastic config (#646)

jeffra committed 4 years ago

Verified bc046dc4

document deepspeed.initialize() (#644)

stas00 committed 4 years ago

Verified 828d75ba

Add deepspeed.init_distributed to RTD page (#645)

jeffra committed 4 years ago

Verified 4e2dc4e4

Bump nokogiri from 1.10.10 to 1.11.0 in /docs (#630)

dependabot[bot] committed 4 years ago

Verified 8cea96dd

Update builder.py (#642)

sxjscience committed 4 years ago

Verified 64461da4

Module replacement support (#586)

jeffra committed 4 years ago

Verified 44bd538b

Fix docstring format (#640)

tjruwase committed 4 years ago

Verified 5ab12795

docs: minor spelling tweaks (#623)

brettkoonce committed 4 years ago

Verified 46d2e287

change dist to torch.distributed to fix bug in assert. (#638)

awan-10 committed 4 years ago

Verified d38ad6a1

Allow DeepSpeed models to be initialized with optimizer=None (#469)

gcooper-isi committed 4 years ago

Verified a9a83a6f

Support initialization with dict configuration (#632)

tjruwase committed 4 years ago

Verified e6ac7311

update SA comp check to fix torch-cpu issue (#631)

jeffra committed 4 years ago

Verified 24e07399

Elastic training support (#602)

jeffra committed 4 years ago

Verified 81aeea36

Ability to initialize distributed backend outside deepspeed runtime (#608)

jeffra committed 4 years ago

Verified 7435b2f1

Transformer-kernel - supporting any arbitrary sequence-length (#587)

RezaYazdaniAminabadi committed 4 years ago

Verified fd2f970b

Fixes for RTD build errors (#606)

jeffra committed 4 years ago

Verified 6380ee35

[doc] xref to hostfile discussion (#604)

stas00 committed 4 years ago

Verified 007466e5

implement missing get_last_lr (#595)

stas00 committed 4 years ago

Verified 9f8e8f38

Update launcher to set local rank environ variable (#597)

jeffra committed 4 years ago

Verified c5a449f9

Supported customizing kwargs for lr_scheduler (#584)

carefree0910 committed 4 years ago

Verified a4763f55

add DeepSpeedZeroConfig repr method (#596)

stas00 committed 4 years ago

Verified 66268bd3

Older