Pull Requests deepspeedai/DeepSpeed

[BUGFIX] Reset `bucket.elements` after reduction in ZeRO Stage 3

#7418 by rahul713rk was merged 2025-07-08 02:32

Update version.txt after v0.17.2 release.

#7417 by loadams was merged 2025-07-07 19:10

Align missing argument in AllReduceCoalescedHandle

#7414 by deepcharm was merged 2025-07-08 21:10

[TiledMLP]: fix for bs>1

#7412 by stas00 was merged 2025-07-07 18:34

fix: engine initializes optimizer attributes at the beginning

#7410 by HollowMan6 was merged 2025-07-07 02:24

Improvements to Communication Logger

#7404 by alexk101 was merged 2025-07-09 17:35

fix: DeepCompile for torch 2.8

#7402 by HollowMan6 was merged 2025-06-30 18:19

fix: Add `csrc/compile` to include paths for DeepCompile builder

#7401 by HollowMan6 was merged 2025-06-30 17:55

fix(comm): Expose GradBucket in deepspeed.comm API

#7400 by Flink-ddd was merged 2025-06-30 20:49

fix: fix FileNotFoundError for build_win.bat

#7399 by gjj2828 was merged 2025-07-01 16:13

Added device detection to communication logging

#7398 by alexk101 was merged 2025-06-28 06:56

Fix unbound local error for `return_val`

#7395 by HollowMan6 was merged 2025-06-27 17:10

add support for CUDAtk12.9

#7394 by loscrossos was merged 2025-06-27 15:24

Add Zenflow code for Stage 1 & 2

#7391 by Antlera was merged 2025-08-15 17:32

fix broken url

#7390 by stas00 was merged 2025-06-26 22:29

Add RISCV64 CPU Support

#7387 by heyujiao99 was closed 2025-07-10 01:05

Improve coverage of DeepCompile

#7386 by tohtana was merged 2025-06-27 23:28

add blog link

#7385 by stas00 was merged 2025-06-24 20:14

fix wandb.log() call by removing `sync` kwarg

#7383 by ned2 was merged 2025-06-23 16:07

Add support for ws=1 scenario

#7379 by NirSonnenschein was merged 2025-06-23 19:07

fix(inference): Add missing dtype attribute to ParameterBase setter

#7378 by Flink-ddd was merged 2025-06-23 23:38

Fix dtype mismatch in `TestParamPartitioningSkipInit`

#7377 by tohtana was merged 2025-06-23 16:44

Fix release of IPG buffer

#7376 by tohtana was merged 2025-06-22 04:24

Update latest news with DeepNVMe

#7375 by loadams was merged 2025-06-20 17:05

Relax tolerances for FP8 unit test only for ROCm + FP16

#7373 by rraminen was merged 2025-06-20 15:24

add Arctic Long Sequence Training paper reference

#7372 by stas00 was merged 2025-06-20 00:25

fix #7188

#7371 by lpnpcs was merged 2025-08-04 18:36

Lower loss_scale on ROCm to enable FP8 unit test

#7367 by rraminen was closed 2025-06-20 17:08

Fix ZeRO stage 1 and add stage 2 support with DeepCompile

#7366 by tohtana was merged 2025-06-27 22:31

Fix tutorial title

#7365 by stas00 was merged 2025-06-17 17:10