Pull Requests deepspeedai/DeepSpeed

Restore full training state in checkpoint writer engines

#8502 opened 2026-09-13 05:54 by ai-yang

Fix ZeRO-3 checkpoint tag selection during weight reconstruction

#8501 opened 2026-09-13 05:53 by ai-yang

docs: add Simplified Chinese translation (README.zh.md)

#8498 opened 2026-09-13 04:30 by JasonYeYuhe

feat(rollout): align decode fronts and trim dead zones

#8497 opened 2026-09-13 02:50 by nathon-lee

Keep ZenFlow Adam step counters in float32

#8496 opened 2026-09-13 00:41 by vineethsaivs

Preserve each Adam group's beta2 in OneCycle

#8495 opened 2026-09-13 00:40 by vineethsaivs

feat(rollout): add continuous batching profiling

#8494 opened 2026-09-12 11:58 by nathon-lee

Deprecate sparse attention

#8493 opened 2026-09-12 05:09 by hwchen2017

Remove triton compatibility check for fp_quantizer

#8492 opened 2026-09-12 00:59 by hwchen2017

[bugfix] _DimZeroAllToAll silently fails to send grads under torch.compile

#8491 opened 2026-09-11 23:19 by stas00

Refresh the whole captured input before a CUDA graph replay

#8488 opened 2026-09-11 19:53 by vineethsaivs

Make wait() idempotent on AllGatherHandle and NoGatherHandle

#8487 opened 2026-09-11 16:37 by pengdurice

tests: allow testing an installed DeepSpeed package

#8485 opened 2026-09-11 12:34 by arubittu

docs: fix precompiled ops install command

#8484 opened 2026-09-11 12:26 by arubittu

Give Muon's momentum the dtype of the gradient it is combined with

#8483 opened 2026-09-11 11:16 by alanhuangyoo

Fixed memory leak in loss backward with AC & zero3 & single-model-multi-branch network

#8482 opened 2026-09-11 10:48 by supermeng

Recognize a norm by its shape, not by whether its class name was listed

#8479 opened 2026-09-10 19:59 by alanhuangyoo

Count each expert once in the fp32 gradient-clipping norm

#8478 opened 2026-09-10 19:07 by alanhuangyoo

Fix AutoEP global L2 clipping for sharded expert ownership

#8476 opened 2026-09-10 16:43 by gss10282023

Reject a non-positive out-token limit at config time

#8473 opened 2026-09-10 14:54 by alanhuangyoo

Make add_tuning_arguments' 1Cycle flags reach OneCycle

#8472 opened 2026-09-10 10:28 by alanhuangyoo

fix: track FusedAdam bias correction per parameter

#8471 opened 2026-09-10 09:36 by MrCapricornLiu

Support dynamic parameter offload for ZeRO-3 inference

#8468 opened 2026-09-10 00:35 by zupengwang

A zero-element parameter loses its shape when it is bound to a flat buffer

#8467 opened 2026-09-09 14:14 by alanhuangyoo

fix(inference): break circular import in ops.transformer.inference

#8465 opened 2026-09-09 02:50 by chakshu-dhannawat

Fix Muon optimizer under ZeRO CPU offload and bound gather buffers

#8464 opened 2026-09-08 16:33 by jinyouzhi

Checkpoint the RNG the curriculum sampler actually draws from

#8460 opened 2026-09-08 12:06 by alanhuangyoo

[AutoSP] Fix compiler correctness

#8457 opened 2026-09-08 10:21 by jinyouzhi

Say when ZenFlow will never re-select its important columns

#8456 opened 2026-09-08 06:45 by alanhuangyoo

Rebuild the full argument list when partitioning activations

#8455 opened 2026-09-08 06:44 by vineethsaivs