deepspeedai/DeepSpeed

Pull Requests Commits

Preserve non-floating buffers in ZeRO checkpoint conversion (#8515)

aprylewu committed 1 day ago

Verified a20acd1d

[UC] Plan a scale-1 affine shard-to-shard transfer without rebuilding the parameter (#8623)

0z5a committed 1 day ago

Verified ae1b183f

Log the skipped optimizer step on gradient overflow (#8417)

sfc-gh-abkulkarni committed 1 day ago

Verified 500997a4

Add an opt-in fused Triton RMSNorm for Llama-style Hugging Face models (#8645)

yh0903 committed 1 day ago

Verified 3ca100d6

AutoEP + AutoTP folding under ZeRO-3 (#8670)

pengdurice committed 1 day ago

Verified 7d77213d

Avoid circuler import caused by Hybrid Engine and OPT model (#8737)

tohtana committed 1 day ago

Verified f47a963d

Reduce unnecessary CI work for documentation changes and consolidate agent instructions in AGENTS.md (#8794)

hwchen2017 committed 1 day ago

Verified bf8f489a

Guard the shm inference all-reduce on the host ISA (#8701)

delock committed 1 day ago

Verified 18f92828

Use default fp32 tolerances in AutoTP training tests (#8796)

hwchen2017 committed 1 day ago

Verified 967803b0

feat(rollout): add OPSD response token loss primitives (#8677)

nathon-lee committed 1 day ago

Verified f1b516b6

Fix DataAnalyzer.run_map crashing when specific_threads is set (#8758)

Rakshit-gen committed 2 days ago

Verified b7f51df9

Keep the buffers a model names for fp32 in fp32 under ZeRO-3 (#8699)

pengdurice committed 2 days ago

Verified 7b911151

[AutoTP] Support tied embeddings for row-parallel output-head training (#8680)

jinyouzhi committed 2 days ago

Verified 3c6a5ca9

Assert offload-state removal only when offload leaves the accelerator (#8738)

delock committed 2 days ago

Verified 1707800a

Stream AWS fallback logs while commands run (#8783)

tohtana committed 2 days ago

Verified ff85fbfc

Release each BF16 gradient once immediate_grad_update has accumulated it (#8654)

yh0903 committed 2 days ago

Verified bc1ad320

[BUG] Replace deprecated `npu_apply_adam(_w)` with `torch._fused_adam(w)_` on NPU (#8640)

Ginray committed 2 days ago

Verified c456c059

Add AWS capacity fallback for torch-latest (#8778)

tohtana committed 2 days ago

Verified 66e106c8

Pin Modal torch-latest to OCI and cover PR open events (#8770)

tohtana committed 3 days ago

Verified c6af3faf

Keep invalid group-norm sentinels from becoming a finite 1.0 step (#8638)

sfc-gh-truwase committed 5 days ago

Verified 991ccf0f

Raise Modal nightly Sandbox timeout to 2h (#8746)

delock committed 5 days ago

Verified b90283d5

Fix autotuner crash when an experiment produces no metric (#8421)

Rakshit-gen committed 5 days ago

Verified f0a3be9b

feat(rollout): align decode fronts and trim dead zones (#8497)

nathon-lee committed 7 days ago

Verified a77aeb67

Add an opt-in fused Triton RoPE for Hugging Face split-half models (#8663)

yh0903 committed 7 days ago

Verified 583fc9c3

Give rollout cache fakes the per-layer config Transformers now reads (#8730)

sfc-gh-truwase committed 8 days ago

Verified 8fb2c9a2

Harden ZeRO-1/2 offload gradient storage lifetime and stream ordering (#8632)

jeffra committed 9 days ago

Verified c687df79

Fix Ulysses SP registration for string model paths and gloo collectives (#8702)

delock committed 9 days ago

Verified 04e0be2b

AutoEP: reorder expert rows with gathers on the collective path (#8671)

yh0903 committed 9 days ago

Verified bfd29175

AutoEP: MiniMax-M3 preset (#8651)

pengdurice committed 9 days ago

Verified b9ce5f42

Add opt-in fused AutoEP DeepEP row weighting (#8644)

yh0903 committed 9 days ago

Verified 2c4ead4b

Older