deepspeedai/DeepSpeed

Pull Requests Commits

Update deepspeed-chat/japanese/README.md (#3765)

eltociear committed 2 years ago

Verified 1491e14e

Fix url in getting-started guide (docs) (#3768)

acforvs committed 2 years ago

Verified 044dd0e2

allocate workspace when base = False

molly-smith committed 2 years ago

29c58a22

Create clean baseline funcs with no prints

molly-smith committed 2 years ago

c04a735e

Add weight copy to base mlp ln, seperate norms from tensors in mlp/attn print statements

lekurile committed 2 years ago

b84287e8

fix ccl_backend and residual_add problems (#3642)

Dino Chen committed 2 years ago

Verified 3f5e4931

Account for expert parameters when calculating the total number of parameters in the model (#3720)

alito committed 2 years ago

Verified 9a8f6a1d

add Chinese Zhihu social account (#3755)

conglongli committed 2 years ago

Verified b692d236

remove UtilsBuilder load, use torch (un)flatten ops (#3728)

inkcherry committed 2 years ago

Verified 5a5340d0

format fix. add print tensors flag.

awan-10 committed 2 years ago

36ba09e5

remove workspace alloc in baseline.

awan-10 committed 2 years ago

12c0e813

Fix output transpose dimension bugs (#3747)

loadams committed 2 years ago

Verified cd911f9a

fix hybrid engine mlp module (#3736)

tensor-tang committed 2 years ago

Verified 45466afa

Include cublas error details when getting cublas handle fails (#3695)

jli committed 2 years ago

Verified 46bb08c2

Fix autotuner get_gas_from_user_config (#3664)

straywarrior committed 2 years ago

Verified 09332dbf

Fix apex install bugs (#3741)

loadams committed 2 years ago

Verified 1b401823

Merge branch 'master' into clean-opt

awan-10 committed 2 years ago

d57eb70e

he with opt-baseline impl. now works.

awan-10 committed 2 years ago

69c5cbfd

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

awan-10 committed 2 years ago

e0402bf0

fix for transposed mode.

awan-10 committed 2 years ago

fc7129c4

FP8 unittest for H100 (#3731)

jomayeri committed 2 years ago

Verified 6f4fc30b

Documentation for DeepSpeed Accelerator Abstraction Interface (#3184)

delock committed 2 years ago

Verified 5289d691

jeffra committed 2 years ago

54bd9e29

Update Dockerfile with newer cuda and torch. (#3716)

loadams committed 2 years ago

Verified a65f6b9e

single node pdsh sigkill (#3730)

abhilash1910 committed 2 years ago

Verified 26b3e732

[Bugfix][CPU] Remove C++ version in CPU OpBuilder (#3643)

delock committed 2 years ago

Verified 8bfbb0e3

Increase tensor creator coverage (#3684)

tjruwase committed 2 years ago

Verified 046afced

Fix typo in name of hybrid engine function (#3704)

loadams committed 2 years ago

Verified fc8e5c88

zero3 performance optimizations (#3622)

BacharL committed 2 years ago

Verified 0977106a

DeepSpeed overview in Japanese (#3709)

conglongli committed 2 years ago

Verified df425097

Newer Older