microsoft/DeepSpeed

Pull Requests Commits

Update asymmetric quant to reduce maximum error at the cost of slightly higher average error

cmikeh2 committed 2 years ago

d7516d28

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

fix some typo (#3675)

digger-yu committed 2 years ago

Verified 3fb3cfdc

[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

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

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

guoyejun committed 2 years ago

Verified 3b299997

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

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

guoyejun committed 2 years ago

Verified 52907a66

AISC launcher fixes (#3637)

jeffra committed 2 years ago

Verified 49a73549

Align InferenceEngine to store ms in _model_times (#3501)

HolyFalafel committed 2 years ago

Verified d755b9d6

Expose Consecutive Hysteresis to Users (#3553)

Quentin-Anthony committed 2 years ago

Verified 0411a9f8

DS init should not broadcast or move zero.Init models (#3611)

tjruwase committed 2 years ago

Verified d39c311f

bug fix (#3609)

conglongli committed 2 years ago

Verified 736bf185

Fix op_builder against PyTorch nightly (#3596)

malfet committed 2 years ago

Verified 6622776c

Fix Hybrid Engine for BLOOM (#3580)

lekurile committed 2 years ago

Verified 76679884

Fixing bf16 test (#3551)

jomayeri committed 2 years ago

Verified 49d399cd

fix_typo (#3559)

inkcherry committed 2 years ago

Verified d1c3c0df

rraminen committed 2 years ago

Verified 2fc56841

Clone tensors to avoid torch.save bloat (#3348)

tjruwase committed 2 years ago

Verified 5c3ebd7e

Older