microsoft/DeepSpeed

Pull Requests Commits

Change source of CPUAdam for xpu accelerator (#5703)

Liangliang-Ma committed 1 year ago

Verified 4b8a4a07

Add an argument to enable the injection of missing state during the conversion of universal checkpoints (#5608)

xylian86 committed 1 year ago

Verified f0e3f01d

Disable nvtx decorator to avoid graph break (#5697)

tohtana committed 1 year ago

Verified b421e8c8

ENV var added for recaching in INF Unit tests (#5688)

raza-sikander committed 1 year ago

Verified e9ffe028

inference unit test injectionPolicy split world_size to multiple tests (#5687)

oelayan7 committed 1 year ago

Verified 4f8531bc

Universal checkpoint for zero stage 3 (#5475)

xylian86 committed 1 year ago

Verified d2b1d7fc

[CPU] add fp16 support to shm inference_all_reduce (#5669)

delock committed 1 year ago

Verified 19da95f7

Fix memory leak from _hp_mapping (#5643)

chiragjn committed 1 year ago

Verified 88b2ef71

Bug fix for the "Link bit16 and fp32 parameters in partition" (#5681)

U-rara committed 1 year ago

Verified 224a05c3

Fixed Windows inference build. (#5609)

costin-eseanu committed 1 year ago

Verified b3767d01

Update version.txt after 0.14.4 release (#5694)

mrwyattii committed 1 year ago

Verified d89e8cdf

[XPU] support op builder from intel_extension_for_pytorch kernel path (#5425)

YizhouZ committed 1 year ago

Verified d254d75e

Update BUFSIZE to come from autotuner's constants.py, not numpy (#5686)

loadams committed 1 year ago

Verified 0c979d67

Fix numpy upgrade to 2.0.0 BUFSIZE import error (#5680)

Yejing-Lai committed 1 year ago

Verified 4000cee8

Fix latest pytorch '_get_socket_with_port' import error (#5654)

Yejing-Lai committed 1 year ago

Verified a4cd5504

enable yuan autotp & add conv tp (#5428)

Yejing-Lai committed 1 year ago

Verified 8ea995ee

Fixing the reshape bug in sequence parallel alltoall, which corrupted all QKV data (#5664)

YJHMITWEB committed 1 year ago

Verified 3bdd187e

Fix hpZ with zero element (#5652)

samadejacobs committed 1 year ago

Verified b33873d2

Remove compile wrapper to simplify access to model attributes (#5581)

tohtana committed 1 year ago

Verified 2a0c0e3c

fix IDEX dependence in xpu accelerator (#5666)

Liangliang-Ma committed 1 year ago

Verified 8831b57b

[XPU] adapt lazy_call func to different versions (#5670)

YizhouZ committed 1 year ago

Verified 73316307

_exec_forward_pass: place zeros(1) on the same device as the param (#5576)

nelyahu committed 1 year ago

Verified 05cb79db

[CPU] SHM based allreduce improvement for small message size (#5571)

delock committed 1 year ago

Verified eda5075b

Update version.txt after 0.14.3 release (#5651)

mrwyattii committed 1 year ago

Verified dfcade24

Monitor was always enabled causing performance degradation (#5633)

deepcharm committed 1 year ago

Verified 54f98fd9

pipe/_exec_backward_pass: fix immediate grad update (#5605)

nelyahu committed 1 year ago

Verified 91a9986d

assumption of torch.initial_seed function accepting seed arg in DeepSpeedAccelerator abstract class is incorrect (#5569)

polisettyvarma committed 1 year ago

Verified ac935c7f

fixes in _partition_param_sec function (#5613)

mmhab committed 1 year ago

Verified b6e24adb

Fix overlap communication of ZeRO stage 1 and 2 (#5606)

penn513 committed 1 year ago

Verified a41729f6

stage_1_and_2: optimize clip calculation to use clamp (#5632)

nelyahu committed 1 year ago

Verified 1ef9b029

Newer Older