microsoft/DeepSpeed

Pull Requests Commits

Fix local rank mismatch for heterogeneous nodes (#3409)

byungsoo-oh committed 2 years ago

Verified b7f463dd

non-JIT build fix on ROCm (#3638)

rraminen committed 2 years ago

Verified 4cd0a003

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

siddharth9820 committed 2 years ago

Verified 2d737edd

Use logger in accelerator (#3682)

tjruwase committed 2 years ago

Verified e5fe5f65

100 percent match between DS (base_mlp + base_attn) and HF.

awan-10 committed 2 years ago

f169851c

Revert "Comment out bias add in attn"

awan-10 committed 2 years ago

1794a144

fix some typo (#3675)

digger-yu committed 2 years ago

Verified 3fb3cfdc

Comment out bias add in attn

molly-smith committed 2 years ago

25870306

remove deepcopy to avoid incompatible torch.jit.script modules.

awan-10 committed 2 years ago

d2b70933

[MiCS] [Fix] saving and loading model checkpoint logic for MiCS sharding (#3440)

zarzen committed 2 years ago

Verified c88af214

jeffra committed 2 years ago

f483c034

Refactor check_enabled root validator in DeepSpeedMonitorConfig (#3616)

bgr8 committed 2 years ago

Verified 4559aa9b

fix typo deepspeed/runtime (#3663)

digger-yu committed 2 years ago

Verified 5d14afd2

flops_profiler: add option recompute_fwd_factor for the case of activation recompute (#3362)

guoyejun committed 2 years ago

Verified 460bec46

fix typo with deepspeed/ (#3547)

digger-yu committed 2 years ago

Verified cd4e473e

Set attention past_layer to presents every pass

molly-smith committed 2 years ago

1fa46ee1

allow dict datatype for checkpoints (#3007)

mrwyattii committed 2 years ago

Verified da8f4e01

Fix RuntimeError when using ZeRO Stage3 with mpu: #3564 (#3565)

eggiter committed 2 years ago

Verified f5dde36c

typo and uncomment mlp bias add

molly-smith committed 2 years ago

dedb3d85

deepspeed/comm/comm.py: fix typo of warning message (#3636)

guoyejun committed 2 years ago

Verified 3b299997

Print more tensors. Add torch syncs to attn.

molly-smith committed 2 years ago

1aed330c

complete attn baseline

molly-smith committed 2 years ago

29415d08

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

awan-10 committed 2 years ago

f746e9df

reorder code to cleanup/merge base_mlp in one place.

awan-10 committed 2 years ago

e245219f

Typo Correction (#3621)

Micah Zoltu committed 2 years ago

Verified e02b8d0b

Update megatron.md (#3641)

wjessup committed 2 years ago

Verified 8f459c50

Skip tests on docs-only changes (#3651)

mrwyattii committed 2 years ago

Verified 8b8c7031

Add Ascend NPU accelerator support (#3595)

CurryRice233 committed 2 years ago

Verified f3c8eaca

Add attention baseline

molly-smith committed 2 years ago

548a3966

stage3.py: do not scale if gradient_predivide_factor is 1.0 (#3630)

guoyejun committed 2 years ago

Verified 52907a66

Newer Older