deepspeedai/DeepSpeed

Pull Requests Commits

squash llama to clean opt (#3801)

awan-10 committed 2 years ago

Verified 928e3fe6

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

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

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

Small tweak on cuda version mismatch documentation (#3706)

jli committed 2 years ago

Verified d414678d

Fix unit test typo in tests/unit/ops/transformer/inference (#3697)

mrwyattii committed 2 years ago

Verified fb2b4ab1

change partititon_name to partition_name (#3700)

digger-yu committed 2 years ago

Verified c5edc91e

Fix gpt-j inference issue (#3639)

RezaYazdaniAminabadi committed 2 years ago

Verified 34a9fbf1

Revert "fix typo name (#3689)" (#3702)

loadams committed 2 years ago

Verified 7e59ef12

enable base impls.

awan-10 committed 2 years ago

d35524f2

cleanup code and put base-impls for opt only.

awan-10 committed 2 years ago

ca2fed2d

fix typo name (#3689)

tensor-tang committed 2 years ago

Verified f2f5f21b

Fix incorrectly formatted f string (#3698)

loadams committed 2 years ago

Verified d8aaa581

Correct world_size/backend for mpi (#3694)

abhilash1910 committed 2 years ago

Verified c17313fb

Older