microsoft/DeepSpeed

Pull Requests Commits

Update with changes from BAcharL's PR

loadams committed 1 year ago

d0574bfb

Un-pin torch version in nv-torch-latest back to latest

loadams committed 1 year ago

681b120c

Update version.txt after 0.14.2 release (#5458)

mrwyattii committed 1 year ago

Verified fbdf0eaf

Update PyTest torch version to match PyTorch latest official (2.3.0) (#5454)

loadams committed 1 year ago

Verified 5f631abc

Update README.md (#5453)

Jhonso7393 committed 1 year ago

Verified ad202795

OptimizedLinear implementation (#5355)

jeffra committed 1 year ago

Verified 5e6c9b93

set the default to use set_to_none for clearing gradients in BF16 optimizer. (#5434)

inkcherry committed 1 year ago

Verified c66bc426

Improve parallel process of universal checkpoint conversion (#5343)

tohtana committed 1 year ago

Verified c292b03a

64bit indexing fused adam (#5187)

garrett4wade committed 1 year ago

Verified 9b6ef9e1

add device config env for the accelerator (#5396)

shiyuan680 committed 1 year ago

Verified 3f875d95

Fix sorting of shard optimizer states files for universal checkpoint (#5395)

tohtana committed 1 year ago

Verified 99951caa

Selective dequantization (#5375)

RezaYazdaniAminabadi committed 1 year ago

Verified c632ea09

Parallel map step for `DistributedDataAnalyzer` map-reduce (#5291)

bm-synth committed 1 year ago

Verified 64defe65

Bug fix in `split_index` method (#5292)

bm-synth committed 1 year ago

Verified aaaf8bc5

Add required paths to trigger AMD tests on PRs (#5406)

loadams committed 1 year ago

Verified 3194fe85

[CPU] Support SHM based inference_all_reduce in TorchBackend (#5391)

delock committed 1 year ago

Verified b22706a7

Add more tests into XPU CI (#5427)

Liangliang-Ma committed 1 year ago

Verified 4d4c99d9

Update _sidebar.scss (#5293)

fasterinnerlooper committed 1 year ago

Verified a9cbd688

Update engine.py to avoid torch warning (#5408)

etiennebonnafoux committed 1 year ago

Verified bc0f7747

Update 'create-pr' action in release workflow to latest (#5415)

loadams committed 1 year ago

Verified 34969d20

Fix a convergence issues in TP topology caused by incorrect grad_norm. (#5411)

inkcherry committed 1 year ago

Verified 0896503e

[XPU] Use non_daemonic_proc by default on XPU device (#5412)

ys950902 committed 1 year ago

Verified 258e500f

Remove dtype(fp16) condition check for residual_add unit test (#5329)

raza-sikander committed 1 year ago

Verified 89491053

Update version.txt after 0.14.1 release (#5413)

mrwyattii committed 1 year ago

Verified 2b3d31f9

Fix the FP6 kernels compilation problem on non-Ampere GPUs. (#5333)

JamesTheZ committed 1 year ago

Verified e3d873a0

stage3: efficient compute of scaled_global_grad_norm (#5256)

nelyahu committed 1 year ago

Verified 54c06872

fix pagable h2d memcpy (#5301)

GuanhuaWang committed 1 year ago

Verified 7b5b0660

Removal of cuda hardcoded string with get_device function (#5351)

raza-sikander committed 1 year ago

Verified f69f8840

Add custom reshaping for universal checkpoint (#5390)

tohtana committed 1 year ago

Verified 2c51aba0

Optimize the fp-dequantizer to get high memory-BW utilization (#5373)

RezaYazdaniAminabadi committed 1 year ago

Verified a8b82153

Older