Skip to content

Pull requests: microsoft/DeepSpeed

New pull request New
146 Open 2,625 Closed
146 Open 2,625 Closed
Author
Filter by author
Label
Filter by label
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Milestones
Filter by milestone
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Monitor was always enabled causing performance degradation
#5633 opened Jun 9, 2024 by deepcharm Loading…
stage_1_and_2: optimize clip calculation to use clamp
#5632 opened Jun 9, 2024 by nelyahu Loading…
reduce all-to-all communication volume when both expert and non-expert are tensor-parallel
#5626 opened Jun 7, 2024 by taozhiwei Loading…
2
Hybrid Offloading for ZeRO3
#5625 opened Jun 7, 2024 by tohtana • Draft
fix: quantization with DeepSpeed HE
#5624 opened Jun 6, 2024 by Atry Loading…
1
Add support for Phi-3 small to FastGen
#5614 opened Jun 4, 2024 by adk9 • Draft
fixes in _partition_param_sec function
#5613 opened Jun 4, 2024 by mmhab Loading…
[INF] Enable torch compile for inference
#5612 opened Jun 4, 2024 by oelayan7 Loading…
2
Upgrade HPU image to v1.16.0.
#5610 opened Jun 4, 2024 by vshekhawat-hlab Loading…
4
Fixed Windows inference build.
#5609 opened Jun 3, 2024 by costin-eseanu Loading…
Add an argument to enable the injection of missing state during the conversion of universal checkpoints
#5608 opened Jun 3, 2024 by xylian86 Loading…
Fix overlap communication of ZeRO stage 1 and 2
#5606 opened Jun 3, 2024 by penn513 Loading…
4
pipe/_exec_backward_pass: fix immediate grad update
#5605 opened Jun 3, 2024 by nelyahu Loading…
[CPU] Allow deepspeed.comm.inference_all_reduce in torch.compile graph
#5604 opened Jun 3, 2024 by delock Loading…
state_dict_factory: llama checkpoint - support SWIGLU
#5601 opened Jun 2, 2024 by nelyahu Loading…
WA for Torch-compile-Z3-act-apt accuracy issue from the Pytorch repo
#5590 opened May 30, 2024 by NirSonnenschein Loading…
5
FastGen H100 MoE support: Add PyTorch multi-gemm MOE implementation
#5586 opened May 29, 2024 by HeyangQin Loading…
7
Update profiler.py
#5584 opened May 29, 2024 by gameofdimension Loading…
Remove compile wrapper to simplify access to model attributes
#5581 opened May 29, 2024 by tohtana Loading…
21
reduce cpu host overhead when using moe
#5578 opened May 29, 2024 by ranzhejiang Loading…
6
_exec_forward_pass: place zeros(1) on the same device as the param
#5576 opened May 28, 2024 by nelyahu Loading…
Reuse KV cache of prefixes
#5572 opened May 27, 2024 by tohtana • Draft
3
[CPU] SHM based allreduce improvement for small message size
#5571 opened May 27, 2024 by delock Loading…
2
assumption of torch.initial_seed function accepting seed arg in DeepSpeedAccelerator abstract class is incorrect
#5569 opened May 26, 2024 by polisettyvarma Loading…
4
Add support for Microsoft Phi-3 model to DeepSpeed-FastGen
#5559 opened May 21, 2024 by adk9 Loading…
ProTip! no:milestone will show everything without a milestone.