microsoft/DeepSpeed

Pull Requests Commits

apply fp16 autocast only to floating point values

Masahiro Tanaka committed 1 year ago

5c4b97f1

Add API to get devices of offload states (#6586)

tohtana committed 1 year ago

Verified adec9912

reduce setting global variables to reduce torch compile graph breaks (#6541)

NirSonnenschein committed 1 year ago

Verified d7ca3d83

AIO CPU Locked Tensor (#6592)

jomayeri committed 1 year ago

Verified a1f98bdc

Clean up prefetched parameters (#6557)

tohtana committed 1 year ago

Verified 7d751ee8

Update version.txt after 0.15.2 release (#6615)

loadams committed 1 year ago

Verified 55f7f378

Enabled Qwen2-MoE Tensor Parallelism (TP) inference (#6551)

gyou2021 committed 1 year ago

Verified 474a3288

Unpin accelerate tests, update lightning with node16 removal. (#6611)

loadams committed 1 year ago

Verified 1062a0c6

Rearrange inference OPS and stop using builder.load (#5490)

oelayan7 committed 1 year ago

Verified 645639bc

Handle when `backend` is also in compile_kwargs (#6502)

oraluben committed 1 year ago

Verified ca8b1fe9

Fix device selection using CUDA_VISIBLE_DEVICES (#6530)

tohtana committed 1 year ago

Verified 5cbbff40

Improve DS logging control (#6602)

tjruwase committed 1 year ago

Verified f74ea69a

Add llama3.2 vision autotp (#6577)

Yejing-Lai committed 1 year ago

Verified e97b4536

Pin accelerate to fix CI failures/issues (#6610)

loadams committed 1 year ago

Verified 745dd48b

Fix SD workflow (#6609)

loadams committed 1 year ago

Verified 00c4b98b

Move V100 workflows from cuda 11.1/11.7 to 12.1 (#6607)

loadams committed 1 year ago

Verified 20695b39

Add SSF Best practices badge (#6604)

loadams committed 1 year ago

Verified 940887de

Cleanup CODEOWNERS file to be valid (#6603)

loadams committed 1 year ago

Verified 239b83a7

[ROCm] Fix subprocess error (#6587)

jagadish-amd committed 1 year ago

Verified b93c7a20

Fix torch include in `op_builder/mlu/fused_adam.py` and update no-torch workflow triggers (#6584)

loadams committed 1 year ago

Verified 8cded575

Fixes on the accelerate side mean we do not need to skip this test (#6583)

loadams committed 1 year ago

Verified 828ddfbb

[COMPILE] workflow for deepspeed + torch.compile (#6570)

YizhouZ committed 1 year ago

Verified d4e18950

add bfloat16 to inference support dtypes (#6528)

nelyahu committed 1 year ago

Verified 1caf6e81

Add APIs to offload states of model, optimizer, and engine (#6011)

tohtana committed 1 year ago

Verified 047bcf6a

[XPU] Support DeepNVMe new code structure (#6532)

Liangliang-Ma committed 1 year ago

Verified d45cfd34

fix errors when setting zero3 leaf modules with torch.compile (#6564)

NirSonnenschein committed 1 year ago

Verified ba58682a

Fix gradient accumulation for Z2+offload (#6550)

tohtana committed 1 year ago

Verified c85c8703

[Accelerator] Cambricon MLU support (#6472)

Andy666G committed 1 year ago

Verified 0fbe96a5

DeepNVMe perf tuning (#6560)

tjruwase committed 1 year ago

Verified a5400974

Use msgpack for p2p comm (#6547)

tohtana committed 1 year ago

Verified 7622cd9e

Older