Skip to content

Pull requests: deepspeedai/DeepSpeed

New pull request New
134 Open 3,711 Closed
134 Open 3,711 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Clamp capacity to num_tokens in MoE gating drop branches (complete #5353)
#8155 opened Jul 18, 2026 by ebarkhordar Loading…
3
Fix repeated gradient scaling across pipeline stages
#8154 opened Jul 18, 2026 by Functionhx Loading…
1
Validate warmup_type in WarmupCosineLR like WarmupLR
#8151 opened Jul 17, 2026 by sohumt123 Loading…
Z3: Support for activation ckpt with frozen params
#8148 opened Jul 16, 2026 by sfc-gh-truwase Collaborator Loading…
6
[AutoTP] Enalbe HF colwise_gather_output to support lm_head replace
#8146 opened Jul 15, 2026 by jinyouzhi Contributor Loading…
6
Fix ZeRO-3 hooks for attribute-delegating modules
#8144 opened Jul 15, 2026 by Functionhx Loading…
1
2
Add governance process page to deepspeed.ai
#8138 opened Jul 13, 2026 by PKUWZP Collaborator Loading…
3 tasks
5
Feature/zero multi loss separate backward
#8135 opened Jul 12, 2026 by nathon-lee Contributor Loading…
10
fix: add zero-guards for divisor/scale edge cases (#7838)
#8133 opened Jul 12, 2026 by liuyun7345 Loading…
7 tasks done
1
ZeRO-3: don't partition frozen params during an activation-checkpoint recompute
#8130 opened Jul 10, 2026 by qgallouedec Contributor Loading…
11
Make DeepCompile ZeRO-3 scheduling memory-aware
#8129 opened Jul 10, 2026 by tohtana Collaborator • Draft
docs: add CHANGELOG.md
#8116 opened Jul 3, 2026 by NovusEdge Contributor Loading…
1
3
Add Ulysses SP support for FLA gated-delta context parallelism
#8114 opened Jul 2, 2026 by xylian86 Collaborator Loading…
4
Fix formatting CI failures in benchmarks/opsd benchmark scripts
#8111 opened Jul 1, 2026 by Copilot AI • Draft
@PKUWZP
1
ZeRO-3: stream partitioning of oversized parameters in zero.Init
#8103 opened Jun 30, 2026 by Achyuthan-S Contributor Loading…
1
11
Warn when zero.Init silently falls back to a single rank (#8084)
#8089 opened Jun 24, 2026 by akshansh47 Loading…
1
8
feat: add Trackio as a new experiment monitoring backend
#8065 opened Jun 15, 2026 by chanduripranav Loading…
1
4
zero3: defer param release during retain_graph backward #7352
#8045 opened Jun 3, 2026 by nathon-lee Contributor • Draft
11
Add Qwen 3.5 preset to AutoTP
#7978 opened Apr 16, 2026 by tohtana Collaborator • Draft
1
1
Refactor/torch autocast encapsulate global state
#7946 opened Apr 2, 2026 by nathon-lee Contributor Loading…
2
Fix ZeRO-3 optimizer initialization validation (#7844)
#7929 opened Mar 28, 2026 by amadhan882 Loading…
1
9
Add torch_xla TPU support for ZeRO-1/2
#7917 opened Mar 21, 2026 by PKUWZP Collaborator Loading…
18
Fix Stage 0 + Ulysses crash: make bwc_tensor_model_parallel_rank() resilient to MP API absence
#7888 opened Mar 6, 2026 by nathon-lee Contributor Loading…
3
fix(zero): Ensure full gradient reduction for Muon optimizer with reduce_scatter
#7878 opened Feb 27, 2026 by nathon-lee Contributor Loading…
1
2
fix: correct DistributedAttention output shape and pad uneven sequence lengths (#7842)
#7868 opened Feb 22, 2026 by harshang03 • Draft
1
1
ProTip! Adding no:label will show everything without a label.