microsoft/DeepSpeed

Pull Requests Commits

Add initial code for tensor logger

lekurile committed 1 year ago

907de758

Fix ds_opt, update prints

lekurile committed 1 year ago

b647e617

Merge branch 'clean-opt' into lekurile/ds_chat_mlp_debug

lekurile committed 1 year ago

52c87a7c

Add comment about zero bias add in base attn + ds mlp

lekurile committed 1 year ago

35311bc8

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

lekurile committed 1 year ago

b84287e8

format fix. add print tensors flag.

awan-10 committed 1 year ago

36ba09e5

remove workspace alloc in baseline.

awan-10 committed 1 year ago

12c0e813

Additional instrumentation

lekurile committed 1 year ago

803cfc20

Merge branch 'debug-ds-inf' into lekurile/ds_chat_mlp_debug

lekurile committed 1 year ago

e96fc03e

Merge branch 'debug-ds-inf' of https://github.com/microsoft/DeepSpeed into debug-ds-inf

lekurile committed 1 year ago

95affb26

additional instrumentation

lekurile committed 1 year ago

57dcc9d7

Merge branch 'master' into clean-opt

awan-10 committed 1 year ago

d57eb70e

he with opt-baseline impl. now works.

awan-10 committed 1 year ago

69c5cbfd

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

awan-10 committed 1 year ago

e0402bf0

fix for transposed mode.

awan-10 committed 1 year ago

fc7129c4

FP8 unittest for H100 (#3731)

jomayeri committed 1 year ago

Verified 6f4fc30b

Documentation for DeepSpeed Accelerator Abstraction Interface (#3184)

delock committed 1 year ago

Verified 5289d691

jeffra committed 1 year ago

54bd9e29

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

loadams committed 1 year ago

Verified a65f6b9e

single node pdsh sigkill (#3730)

abhilash1910 committed 1 year ago

Verified 26b3e732

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

delock committed 1 year ago

Verified 8bfbb0e3

Increase tensor creator coverage (#3684)

tjruwase committed 1 year ago

Verified 046afced

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

loadams committed 1 year ago

Verified fc8e5c88

zero3 performance optimizations (#3622)

BacharL committed 1 year ago

Verified 0977106a

DeepSpeed overview in Japanese (#3709)

conglongli committed 1 year ago

Verified df425097

Small tweak on cuda version mismatch documentation (#3706)

jli committed 1 year ago

Verified d414678d

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

mrwyattii committed 1 year ago

Verified fb2b4ab1

change partititon_name to partition_name (#3700)

digger-yu committed 1 year ago

Verified c5edc91e

Fix gpt-j inference issue (#3639)

RezaYazdaniAminabadi committed 1 year ago

Verified 34a9fbf1

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

loadams committed 1 year ago

Verified 7e59ef12

Older