Pull Requests microsoft/DeepSpeed

Include init file for superoffload folder

#7591 by nguyen599 was merged 2025-09-24 16:50

Deepcompile: Make size of activation to free configurable

#7582 by eternalNight was merged 2025-09-24 01:37

Broadcast fp16 overflow in Z1

#7580 by sfc-gh-truwase was merged 2025-09-23 15:51

logging: Also set log level of logger handlers

#7576 by eternalNight was merged 2025-09-23 03:32

Update version.txt post 0.17.6 release

#7572 by loadams was merged 2025-09-19 21:33

Deepcompile: Fix bugs when applying deepcompile to VLA-like models

#7569 by eternalNight was merged 2025-09-23 07:27

[bugfix] fix partition context unpatch

#7566 by hjh0119 was merged 2025-09-19 07:24

deepcompile: Create dummy inputs using empty_strided

#7564 by eternalNight was merged 2025-09-15 21:19

deepcompile: Record graph order using OrderedDict

#7563 by eternalNight was merged 2025-09-16 05:25

deepcompile: Create a full list of no-copy ops

#7562 by eternalNight was merged 2025-09-16 16:05

scripts: Check .is_cuda only in non-C++ files

#7561 by eternalNight was merged 2025-09-19 05:01

fix npu device_id AttributeError issue

#7560 by we1sper was merged 2025-09-17 07:46

SuperOffload Release

#7559 by xylian86 was merged 2025-09-24 13:09

Add dependency for deepcompile test

#7558 by tohtana was merged 2025-09-13 07:45

Move modal tests to tests/v1

#7557 by tohtana was merged 2025-09-12 21:28

Make Muon optimizer easier to enable

#7555 by delock was merged 2025-09-17 13:52

Limit random seed range in tests

#7553 by tohtana was merged 2025-09-10 17:45

[MoE] Fix misuse of num_experts as expert parallel group size (ep_size)

#7551 by Flakes342 was merged 2025-09-10 05:31

[ALST tutorial] support bs>1

#7550 by sfc-gh-sbekman was merged 2025-09-09 19:51

Fix gradient buffer access for DeepCompile Z1/2

#7548 by tohtana was merged 2025-09-10 18:12

Improve error message and reduce validation in autocast test

#7547 by tohtana was merged 2025-09-05 07:04

Revert "Add index to HPU devices (#7497)"

#7545 by deepcharm was merged 2025-09-08 22:07

Fix zenflow_torch_adam.py

#7544 by stas00 was merged 2025-09-03 18:14

Relax restrictions of torch.autocast integration

#7543 by tohtana was merged 2025-09-03 19:15

avoid setting device_id to `init_process_group`

#7542 by kaixuanliu was merged 2025-09-05 06:06

[Deepspeed]Fix Muon optimizer import error

#7539 by pengdurice was closed 2025-09-02 18:23

[doc] fixing moe tutorial

#7538 by stas00 was merged 2025-09-02 20:53

[MoE] Fix misuse of num_experts as expert parallel group size (ep_size)

#7537 by Flakes342 was closed 2025-09-09 21:07

undo the revert

#7536 by stas00 was merged 2025-09-02 18:24

Fix scaling and allgather with `torch.autocast`

#7534 by tohtana was merged 2025-09-03 01:22