microsoft/DeepSpeed

Pull Requests Commits

Support bf16 optimizer states with CPU offload (#8010)

lucaspirola committed 171 days ago

Verified 3c337b54

Make GitHub Actions job names unique (#8014)

tohtana committed 171 days ago

Verified 9f95452e

Run FastFileWriter fd-close test outside pytest-forked (#8015)

tohtana committed 171 days ago

Verified d6d4b3d0

Add configurable torch-latest dependency versions (#8016)

tohtana committed 171 days ago

Verified b387166f

Remove stale step() docstring from DeepSpeedCPUAdam (#8011)

lucaspirola committed 172 days ago

Verified ae075141

Fix ZeRO-3 forward crash on modules with plain dict _parameters (#8009)

roycho96 committed 172 days ago

Verified d7a3972f

fix: use subprocess instead of os.system in data_analyzer.py (#7994)

orbisai0security committed 172 days ago

Verified 8cdf8651

Auto-detect CUTLASS for EvoformerAttention (#8000)

MaxTretikov committed 173 days ago

Verified d5356e07

fix(io): close aio_fd in FastFileWriter._fini to prevent fd leak (#8005)

jg-heo committed 174 days ago

Verified b01a0915

zero3: SDMA allgather via mori (sdma_allgather) (#7999)

inkcherry committed 177 days ago

Verified 66af8f03

Optimize singleton MoE collectives (#7997)

Tianyi-Franklin-Wang committed 178 days ago

Verified 55491413

Update topkgating probability-mask test expectation (#8007)

tohtana committed 178 days ago

Verified 5be4c70c

Add office hours times/link on the README (#8004)

loadams committed 180 days ago

Verified 883fefb2

Update version post-v0.19.0 release (#7996)

loadams committed 184 days ago

Verified a7811998

Update version.txt before 0.19.0 release (#7995)

loadams committed 185 days ago

Verified 98a2fbdf

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

nathon-lee committed 190 days ago

Verified 4e668fce

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

delock committed 191 days ago

Verified 8a77f381

Add DeepSpeed NVTX domain support (#7988)

heurry committed 191 days ago

Verified 5b441027

fix: topkgating major bug (#7986)

excepshenal committed 191 days ago

Verified 853c9389

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

maxyu1115 committed 192 days ago

Verified 5999fb06

Fix modal ci timeout (#7989)

sfc-gh-truwase committed 192 days ago

Verified 6e291bee

Dynamic offload compatible with static optimizer offload (#7979)

sfc-gh-truwase committed 197 days ago

Verified f6d7ae48

Fix Adam subgroup inconsistency (#7982)

st-bang97 committed 198 days ago

Verified 44c51e34

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

eternalNight committed 199 days ago

Verified 077bff55

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

roycho96 committed 199 days ago

Verified aeb10bb1

Avoid CUDA reinit error in CI tests (#7977)

tohtana committed 204 days ago

Verified 83d43c35

Rename dequantization template parameters (#7976)

Flamefire committed 206 days ago

Verified dc0fd295

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

Cursx committed 206 days ago

Verified 893c6d20

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

Cursx committed 206 days ago

Verified 3ba3dc97

Enable PyTorch version selection for full test (#7968)

tohtana committed 209 days ago

Verified 0ba23529

Newer Older