microsoft/DeepSpeed

Pull Requests Commits

Move print statements to debug versions

lokoppakmsft committed 2 years ago

5be75dc2

functional clean llama

lokoppakmsft committed 2 years ago

b1507c16

initial prints for attn

molly-smith committed 2 years ago

db0e00a2

allocate workspace when base = False

molly-smith committed 2 years ago

95dc3da2

Create clean baseline funcs with no prints

molly-smith committed 2 years ago

5e0f40b6

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

lekurile committed 2 years ago

69f3ce48

format fix. add print tensors flag.

awan-10 committed 2 years ago

ac3eee13

remove workspace alloc in baseline.

awan-10 committed 2 years ago

6fed0a49

FP8 unittest for H100 (#3731)

jomayeri committed 2 years ago

c1214479

Documentation for DeepSpeed Accelerator Abstraction Interface (#3184)

delock committed 2 years ago

59e4aaa4

jeffra committed 2 years ago

8b7423d2

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

loadams committed 2 years ago

3b161f5e

single node pdsh sigkill (#3730)

abhilash1910 committed 2 years ago

15e9b94c

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

delock committed 2 years ago

25f95ded

Increase tensor creator coverage (#3684)

tjruwase committed 2 years ago

4f9f7bf8

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

loadams committed 2 years ago

8276c61e

zero3 performance optimizations (#3622)

BacharL committed 2 years ago

9ebeaec7

DeepSpeed overview in Japanese (#3709)

conglongli committed 2 years ago

4b8a15c5

Small tweak on cuda version mismatch documentation (#3706)

jli committed 2 years ago

3884a55c

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

mrwyattii committed 2 years ago

0c570d9c

change partititon_name to partition_name (#3700)

digger-yu committed 2 years ago

43697306

Fix gpt-j inference issue (#3639)

RezaYazdaniAminabadi committed 2 years ago

a82f6449

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

loadams committed 2 years ago

1e9af888

fix typo name (#3689)

tensor-tang committed 2 years ago

a1ec835c

Fix incorrectly formatted f string (#3698)

loadams committed 2 years ago

a789a8c6

Correct world_size/backend for mpi (#3694)

abhilash1910 committed 2 years ago

be065644

Fix local rank mismatch for heterogeneous nodes (#3409)

byungsoo-oh committed 2 years ago

4224015c

non-JIT build fix on ROCm (#3638)

rraminen committed 2 years ago

829d3334

Update README to add ICS'23 paper (#3687)

siddharth9820 committed 2 years ago

5b1e1637

Use logger in accelerator (#3682)

tjruwase committed 2 years ago

0d21eeac

Older