microsoft/DeepSpeed

Pull Requests Commits

Merge branch 'master' into fix-sp-dense

mrwyattii committed 2 years ago

Verified bff46e51

[AutoTP] Make AutoTP work when num_heads not divisible by number of workers (#4011)

delock committed 2 years ago

Verified f15cccfa

Enable universal checkpoint for zero stage 1 (#4516)

tjruwase committed 2 years ago

Verified 8fdd9b35

samadejacobs committed 2 years ago

568ae5a6

Allow users to set/override sp comm data type from ds config

samadejacobs committed 2 years ago

01ccf331

Add missing RocBlas include (#4557)

loadams committed 2 years ago

Verified 869629c2

pipe engine _aggregate_total_loss: more efficient loss concatenation (#4327)

nelyahu committed 2 years ago

Verified a02de228

Fix RTD builds (#4558)

mrwyattii committed 2 years ago

Verified 0f2338f7

Merge branch 'master' into fix-sp-dense

tjruwase committed 2 years ago

Verified 7ae577cd

ROCm 6.0 prep changes (#4537)

loadams committed 2 years ago

Verified e2383511

[docs] paper updates (#4543)

jeffra committed 2 years ago

Verified 488f7e2d

turn off I_MPI_PIN for impi launcher (#4531)

delock committed 2 years ago

Verified 04cd6af1

Switch from HIP_PLATFORM_HCC to HIP_PLATFORM_AMD (#4539)

loadams committed 2 years ago

Verified c7724c61

[docs] ZeRO infinity slides and blog (#4542)

jeffra committed 2 years ago

Verified a5b1cb1e

Added rocblas header (#4538)

rraminen committed 2 years ago

Verified 3e4a5871

[Bug fix] Add rope_theta for llama config (#4480)

cupertank committed 2 years ago

Verified beed962c

Fixed deepspeed.comm.monitored_barrier call (#4496)

Quentin-Anthony committed 2 years ago

Verified 8a93ded8

Masahiro Tanaka committed 2 years ago

aaae9949

fix comm group for allreduce

Masahiro Tanaka committed 2 years ago

0bb95947

fix error type in ccl.py (#4521)

Liangliang-Ma committed 2 years ago

Verified a7358817

fix the gradient scale for when zero is not enabled

Reza Yazdani committed 2 years ago

8d901bfc

Add NPU FusedAdam support (#4343)

CurryRice233 committed 2 years ago

Verified 3e70a887

clear redundant parameters in zero3 bwd hook (#4520)

inkcherry committed 2 years ago

Verified 2cbfb89a

fix the sequence-parallelism for the dense models

Reza Yazdani committed 2 years ago

066644d7

add available memory check to accelerators (#4508)

jeffra committed 2 years ago

Verified 12aedac6

Update README.md (#4518)

samadejacobs committed 2 years ago

Verified 78c518ed

Ulysses: add col-ai evaluation (#4517)

samadejacobs committed 2 years ago

Verified 5bbbf41f

[CCLBackend] update API (#4378)

Liangliang-Ma committed 2 years ago

Verified 4fc181b0

add DeepSpeed4Science white paper (#4502)

conglongli committed 2 years ago

Verified 574fbc0d

Fix scale factor on flops profiler (#4500)

loadams committed 2 years ago

Verified a25a67a0

Older