-
Notifications
You must be signed in to change notification settings - Fork 5k
Pull requests: deepspeedai/DeepSpeed
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Preallocate the static KV cache with config.head_dim
#8389
opened Sep 2, 2026 by
ebarkhordar
Contributor
Loading…
Offload a saved view when it is the last value holding its storage
#8388
opened Sep 1, 2026 by
pengdurice
Contributor
•
Draft
Use the val argument in mask_nan_or_inf_with_val_inplace
#8386
opened Sep 1, 2026 by
vineethsaivs
Contributor
Loading…
Describe universal checkpoint shards as affine maps
#8385
opened Sep 1, 2026 by
Achyuthan-S
Contributor
Loading…
[muon] Per-head Newton-Schulz for attention projections
#8384
opened Sep 1, 2026 by
alanhuangyoo
Contributor
Loading…
Add nccl_version to the source-checkout torch_info fallback
#8383
opened Sep 1, 2026 by
alanhuangyoo
Contributor
Loading…
Fix silent gradient-bucket drop on CPU in ZeRO-1/2 non-contiguous reduction
#8382
opened Sep 1, 2026 by
delock
Collaborator
Loading…
Run multi-rank CPU unit tests in CI via LOCAL_SIZE
#8381
opened Sep 1, 2026 by
delock
Collaborator
Loading…
[AutoEP] Make router weight persistent to avoid 0 tensor size crash
#8379
opened Aug 31, 2026 by
pengdurice
Contributor
Loading…
Return frozen parameters as fp32 from get_fp32_state_dict_from_zero_checkpoint
#8378
opened Aug 31, 2026 by
vineethsaivs
Contributor
Loading…
[AutoEP]Fix optimizer and replaced MOE parameter mismatch
#8377
opened Aug 31, 2026 by
pengdurice
Contributor
•
Draft
Fix zero_to_fp32 --safe_serialization on models with tied weights
#8376
opened Aug 31, 2026 by
vineethsaivs
Contributor
Loading…
Gather zero-sized parameters in ZeRO-3
#8375
opened Aug 31, 2026 by
alanhuangyoo
Contributor
Loading…
Build the inference rotary embedding through LlamaConfig
#8373
opened Aug 31, 2026 by
alanhuangyoo
Contributor
Loading…
docs: add test discipline rules to agent guidelines
#8372
opened Aug 31, 2026 by
delock
Collaborator
Loading…
feat(rollout): add continuous batching generation prototype
#8368
opened Aug 30, 2026 by
nathon-lee
Contributor
•
Draft
ZeRO-1/2: synchronize IPG buffer reuse per buffer
#8366
opened Aug 30, 2026 by
almightychang
Loading…
fix(inference): validate keep_module_on_host mode
#8363
opened Aug 30, 2026 by
FU-max-boop
Contributor
Loading…
Reshape instead of view in TiledFusedLogitsLoss
#8362
opened Aug 30, 2026 by
alanhuangyoo
Contributor
Loading…
Reuse existing hpZeRO group in nested zero.Init contexts
#8361
opened Aug 30, 2026 by
zupengwang
Contributor
Loading…
Fix the dead grad_accum branch in the ZenFlow gradient copy
#8360
opened Aug 29, 2026 by
vineethsaivs
Contributor
Loading…
Previous Next
ProTip!
Type g i on any issue or pull request to go back to the issue listing page.