deepspeedai/DeepSpeed

Pull Requests Commits

test: move multimodal tests to tests/unit/v1/multimodal

delock committed 3 days ago

60f1a0c3

test: guard Gemma4Config import with version check

delock committed 7 days ago

82292421

test: add Gemma4 multimodal config test

delock committed 7 days ago

cb84e9c4

fix: use get_text_config() for Gemma4 num_attention_heads fallback

delock committed 7 days ago

635a90db

[AutoSP] (Sequence Parallelism) support for Multimodal Models (ViT + LLM) (#7984)

nathon-lee committed 8 days ago

Verified 4e668fce

Add Gram Newton-Schulz orthogonalization for Muon optimizer (#7953)

delock committed 9 days ago

Verified 8a77f381

Add DeepSpeed NVTX domain support (#7988)

heurry committed 9 days ago

Verified 5b441027

fix: topkgating major bug (#7986)

excepshenal committed 9 days ago

Verified 853c9389

Fix BF16_Optimizer last-microbatch grad leak under ZeRO-1 (#7985)

maxyu1115 committed 10 days ago

Verified 5999fb06

Fix modal ci timeout (#7989)

sfc-gh-truwase committed 11 days ago

Verified 6e291bee

Dynamic offload compatible with static optimizer offload (#7979)

sfc-gh-truwase committed 15 days ago

Verified f6d7ae48

Fix Adam subgroup inconsistency (#7982)

st-bang97 committed 16 days ago

Verified 44c51e34

deepcompile: Fix backward graph recompilation due to unbalanced forward/backward visits (#7980)

eternalNight committed 17 days ago

Verified 077bff55

Fix ZeRO-1/2 CPU-offloaded gradient loss with multiple backward() per step (#7981)

roycho96 committed 17 days ago

Verified aeb10bb1

Avoid CUDA reinit error in CI tests (#7977)

tohtana committed 22 days ago

Verified 83d43c35

Rename dequantization template parameters (#7976)

Flamefire committed 24 days ago

Verified dc0fd295

fix(op_builder): avoid duplicate/wrong -gencode flags (#7974)

Cursx committed 24 days ago

Verified 893c6d20

fix(fp_quantizer): fix UB and negative shift warnings in fp_quantize_impl.cu (#7973)

Cursx committed 24 days ago

Verified 3ba3dc97

Enable PyTorch version selection for full test (#7968)

tohtana committed 27 days ago

Verified 0ba23529

Fix WarmupCosineLR multi-group initialization (#7969)

tohtana committed 27 days ago

Verified 3fd762cf

Fix DeepCompile+Z3 on PyTorch v2.9/2.10 (#7951)

tohtana committed 28 days ago

Verified ecb26a51

Fix overlap-comm buffer lifetimes (#7965)

tohtana committed 28 days ago

Verified dac1525b

Fix zero 1 and 2 CPU-offloaded gradient norm (#7967)

alek6kun committed 29 days ago

Verified 9d632f10

Fix FPQuantizer build (#7963)

Flamefire committed 31 days ago

Verified 3a5df512

fix(zero): detach flat buffer to prevent autograd inplace error on CP… (#7948)

delock committed 36 days ago

Verified 37e232fa

[SP] add SP deny list instead of allow (#7887)

kashif committed 38 days ago

Verified bf0126b5

Zero3 defragment utility (#7940)

nathon-lee committed 39 days ago

Verified 5dce124e

Fix process hang in process-group shutdown (#7941)

Flamefire committed 39 days ago

Verified 2f0924a5

Fix/fix autotp universal checkpoint ci (#7937)

tohtana committed 39 days ago

Verified 3bdebc03

Refactor consolidate transpose (#7934)

nathon-lee committed 40 days ago

Verified 89bf0d2d

Older