microsoft/DeepSpeed

Pull Requests Commits

Add more changes to support CUDA-Graph

Reza Yazdani committed 3 years ago

93562a6b

initial changes for CUDA-Graph support

Reza Yazdani committed 3 years ago

f3b3d491

remove the print line in _einsum_flops_compute function (#1885)

VisionTheta committed 3 years ago

Verified d8ed3ce4

add now required `-lcurand` to solve `undefined symbol: curandCreateGenerator` (#1879)

stas00 committed 3 years ago

Verified c4873722

[docs] fix dead links (#1877)

jeffra committed 3 years ago

Verified 0149bd4f

Supporting multiple modules injection with a single policy when they have identical architectures (#1869)

samyam committed 3 years ago

Verified c13457b7

[docs] add moe paper (#1875)

jeffra committed 3 years ago

Verified 4a356d08

[docs] add amd blog to website (#1874)

jeffra committed 3 years ago

Verified b9e15296

Fix broken links (#1873)

tjruwase committed 3 years ago

Verified d6db6510

Add support for AWS SageMaker. (#1868)

matherit committed 3 years ago

Verified b47e25bf

Fix setup.py crash when torch is not installed. (#1866)

PaperclipBadger committed 3 years ago

Verified 73b5d983

jeffra committed 3 years ago

398f0603

qkv_out can be a single tensor or a list. Handling these cases separetely. (#1850)

samyam committed 3 years ago

Verified ebbcfd52

Import ABC from collections.abc for Python 3.10 compatibility. (#1851)

tirkarthi committed 3 years ago

Verified c7af747c

fix dead MoQ link (#1855)

jeffra committed 3 years ago

Verified 208d45bb

Update config-json.md (#1853)

xldrx committed 3 years ago

Verified b61d7199

deepscale --> deepspeed in prints. (#1854)

awan-10 committed 3 years ago

Verified 788e1c40

add amd release blog (#1848)

jeffra committed 3 years ago

Verified 0eb2c763

Add concurrency policy to CI workflow (#1844)

mrwyattii committed 3 years ago

Verified 2e1847d6

Track only trainable parameters (#1780)

tjruwase committed 3 years ago

Verified b84edef2

fix: Fix undefined variable in _create_expert_data_and_model_parallel and make it easier to understand (#1826)

shjwudp committed 3 years ago

Verified 1e61c7a8

fix: fix undefined variable in MoE top2gating (#1827)

shjwudp committed 3 years ago

Verified 5fb4256a

THCGeneral.h header file is deprecated (#1842)

rraminen committed 3 years ago

Verified b4e8f18c

[ZeRO-1] fix bug w. cpu-offload + > 1 GPU (#1841)

jeffra committed 3 years ago

Verified 28434c00

[build] support cuda-11.6 (#1836)

stas00 committed 3 years ago

Verified 18ea8b79

Split github action unit tests by platform/GPU (#1828)

mrwyattii committed 3 years ago

Verified 41d90830

[launcher] validate passwordless-ssh works when using hostfile launching (#1832)

jeffra committed 3 years ago

Verified a773996d

Reduce specified gradients (#1818)

tjruwase committed 3 years ago

Verified 89b8c887

Tune webpage width (#1829)

cli99 committed 3 years ago

Verified 8a35daf0

[website] remove extra news header

jeffra committed 3 years ago

Verified 9bc296e1

Older