Skip to content

Pull requests: deepspeedai/DeepSpeed

New pull request New
113 Open 3,466 Closed
113 Open 3,466 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

Fix loss scaling and backward call of ZenFlow
#7793 opened Jan 18, 2026 by tohtana Loading…
Fix bf16 dtype mismatch in ZeRO-3 with zero_quantized_weights
#7792 opened Jan 18, 2026 by juyterman1000 Loading…
1
Add bf16 model with fp32 grad_accum to supported configs
#7790 opened Jan 18, 2026 by tohtana Loading…
Skip empty parameters in gradient reduction
#7789 opened Jan 18, 2026 by tohtana Loading…
2
Fix issue with BF16 optimizer selection
#7788 opened Jan 18, 2026 by tohtana Loading…
Fix backward for pipeline engine
#7787 opened Jan 18, 2026 by tohtana Loading…
fix checkpointing/loading of z0+bf16
#7786 opened Jan 17, 2026 by tohtana Loading…
Add workflow to run full tests
#7783 opened Jan 17, 2026 by tohtana • Draft
Fix Muon optimizer conflict with gradient clipping in ZeRO 1/2
#7776 opened Jan 12, 2026 by fy817 Loading…
1
Fix: ZenFlow Adam integration for updated PyTorch backward flow (#7759)
#7771 opened Jan 11, 2026 by Antlera Loading…
9
Introduce all_reduce_hook to support gradient aggregation across replica groups.
#7764 opened Jan 7, 2026 by zhengchenyu Loading…
5
feat: add parameter-level precision control for BF16 training
#7750 opened Dec 30, 2025 by nathon-lee Loading…
1
1
Fix Muon optimizer checkpoint resume with bf16 mode
#7748 opened Dec 28, 2025 by yurekami Loading…
2 tasks done
1
Introduce Megatron-style parallel state management
#7726 opened Dec 15, 2025 by eternalNight • Draft
1 of 5 tasks
3
let allgather and alltoall execute in parallel when both attention and MOE used TP
#7723 opened Dec 11, 2025 by taozhiwei Loading…
2
fix: When there are tensors registered with register buffer in the weight file, the weights are only loaded on device 0 when loading weights across multiple devices.
#7717 opened Dec 8, 2025 by KeeProMise Loading…
if no expert found in parameter that have expert in name the loop should continue
#7685 opened Nov 11, 2025 by LckyLke • Draft
4
Add sequential allgather optimization for ZeRO-3
#7661 opened Oct 31, 2025 by aeeeeeep Loading…
13
Configures workflow for offline unit tests
#7512 opened Aug 24, 2025 by porfanid Loading…
@loadams
6
Add world-size getter in Engine
#7479 opened Aug 9, 2025 by WoosungMyung Loading…
5
Add EXAONE 4.0 model support for DeepSpeed inference v2 @
#7456 opened Jul 29, 2025 by notkisk • Draft
6
Create COMMITTERS_RESPONSIBILITY.md
#7300 opened May 21, 2025 by PKUWZP Loading…
8
HF2UCP: Converting a pytorch_model.bin or .safetensors checkpoint to UCP
#7212 opened Apr 10, 2025 by Schwidola0607 Loading…
17
gather output layout support for column parallel
#7181 opened Mar 28, 2025 by inkcherry Loading…
3
[bugfix] update results of state_dict loading, embedding resizing to secondary partitions (hpz)
#7130 opened Mar 11, 2025 by cyr0930 Loading…
6
ProTip! no:milestone will show everything without a milestone.