microsoft/DeepSpeed

Pull Requests Commits

__del__ is not good for zero-inference

jeffra committed 2 years ago

127e8d35

fix memory leak with z3

jeffra committed 2 years ago

516009b4

fix "undefined symbol: curandCreateGenerator" for quantizer op (#3846)

jinzhen-lin committed 2 years ago

Verified 7e8bcc07

[profiling][mics]Fix some issues for log_summary(). (#3899)

ys950902 committed 2 years ago

Verified c520d476

[ROCm] Enable TestCUDABackward::test_backward unit tests (#3849)

rraminen committed 2 years ago

Verified d24629f4

Fix a typo of global variable in comm.py(#3852) (#3852)

hipudding committed 2 years ago

Verified c5e55d3d

avoid init for deepspeed backend first (#3893)

mrwyattii committed 2 years ago

Verified b58e0fa9

Avoid deprecation warnings in `CHECK_CUDA` (#3854)

Flamefire committed 2 years ago

Verified 9aeba94a

Update workflows for merge queue (#3892)

mrwyattii committed 2 years ago

Verified 52844f49

fix rnn flop profiler to compute flops instead of macs (#3833)

pinstripe-potoroo committed 2 years ago

Verified 3491e32d

fix error :Dictionary expression not allowed in type annotation Pylance (#3708)

digger-yu committed 2 years ago

Verified 97e7b841

Fix Meta Tensor checkpoint load for BLOOM models (#3885)

lekurile committed 2 years ago

Verified cc3a7c9c

Add GPTNeoX AutoTP support (#3778)

Yejing-Lai committed 2 years ago

Verified d6f62217

Separate ZeRO3 InflightParamRegistry for train and eval (#3884)

HeyangQin committed 2 years ago

Verified 9377921a

Extend HE-Lora test with Z3 support + Fix/add guard in HE for Z3 (#3883)

awan-10 committed 2 years ago

Verified db4638d1

Update apex installation to resolve apex's pyproject.toml issues. (#3745)

loadams committed 2 years ago

Verified 59c9b091

Add FALCON Auto-TP Support (#3640)

RezaYazdaniAminabadi committed 2 years ago

Verified f3c93b05

update MMEngine support in README.md (#3879)

HAOCHENYE committed 2 years ago

Verified 385e89d4

update lightning version in CI (#3882)

mrwyattii committed 2 years ago

Verified 7c126f43

Fix LoRA Fuse/Unfuse in Hybrid Engine (#3563)

sxjscience committed 2 years ago

Verified d81dfdab

fix retrieval of out_channels in _conv_trans_flops_compute (#3834)

pinstripe-potoroo committed 2 years ago

Verified c1c1d249

[zero] revert PR #3166, it disabled grad clip for bf16 (#3790)

jeffra committed 2 years ago

Verified 691d246e

Zero3 Fix allreduce optimization for extra large tensor (#3832)

BacharL committed 2 years ago

Verified d229ff17

scripts/check-torchcuda.py: add checking for tensor.is_cuda (#3843)

guoyejun committed 2 years ago

Verified 807d1b5d

checking process_group before merging bucket ranges (#3521) (#3577)

Alexander Jipa committed 2 years ago

Verified 2ded2ff0

[profiling]add show_straggler argument to log_summary() (#3579)

delock committed 2 years ago

Verified 5d1124f2

Reduce Unit Test Time (Part 2) (#3838)

mrwyattii committed 2 years ago

Verified fd1d2c64

Disable AMD test flows in YML (#3847)

loadams committed 2 years ago

Verified c973e157

zero/mics.py: use on_accelerator instead of cuda only (#3806)

guoyejun committed 2 years ago

Verified b4626194

Fix racing condition in GatheredParameters (#3819)

HeyangQin committed 2 years ago

Verified f8551b43

Older