Skip to content

Pull requests: deepspeedai/DeepSpeed

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
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort

Pull requests list

fix: preserve CUDA architecture variant suffixes
#8619 opened Sep 21, 2026 by adenzhou1350 Loading…
Add device-agnostic ZeRO-3 CPU-offload end-to-end benchmark
#8618 opened Sep 21, 2026 by delock Collaborator Loading…
fix: Remove the unused ROCm Apex clone, build, and PYTHONPATH setup.
#8617 opened Sep 21, 2026 by nathon-lee Contributor Loading…
Remove stale compression test from HPU CI
#8616 opened Sep 21, 2026 by nathon-lee Contributor Loading…
Say why a bf16 gradient norm cannot be clipped from
#8614 opened Sep 20, 2026 by vineethsaivs Contributor Loading…
Include the expert gradients in the unfused fp16 clip norm
#8613 opened Sep 20, 2026 by vineethsaivs Contributor Loading…
Forward the per-head Muon tag at every muon_update call site
#8612 opened Sep 20, 2026 by vineethsaivs Contributor Loading…
Fix pipe tests failures on CPU
#8611 opened Sep 20, 2026 by jinyouzhi Contributor Loading…
Skip fp16 ZeroPP tests on accelerators without fp16 support
#8610 opened Sep 20, 2026 by jinyouzhi Contributor Loading…
Follow the optimizer's own state names in universal checkpoints
#8609 opened Sep 20, 2026 by alanhuangyoo Contributor Loading…
Implement graph operations for NPU accelerator
#8605 opened Sep 20, 2026 by VenusTZZ Contributor Loading…
Clear the gradient ZenFlow just offloaded, not param.grad
#8601 opened Sep 20, 2026 by vineethsaivs Contributor Loading…
docs: fix two dead 1Cycle tutorial links
#8593 opened Sep 19, 2026 by simpleqt Loading…
ProTip! Updated in the last three days: updated:>2026-09-18.