microsoft/DeepSpeed

Pull Requests Commits

Update opbuilder

loadams committed 236 days ago

9b72d661

Add defence for DeepCompile w/o optimizer (#7225)

HollowMan6 committed 236 days ago

Verified 86e51e69

Fix build on AMD GPUs (related to DeepCompile) (#7224)

HollowMan6 committed 236 days ago

Verified 0f224d79

Fix pass for z3 and profiler (#7222)

tohtana committed 236 days ago

Verified 1d022e05

Fix release links (#7219)

tjruwase committed 237 days ago

Verified 8f93f8b9

Update version.txt after 0.16.6 release (#7218)

loadams committed 237 days ago

Verified a01a2688

DeepCompile for enhanced compiler integration (#7154)

tohtana committed 237 days ago

Verified 227a60c0

Add defence for offload_states and reload_states w/o optimizer (#7211)

HollowMan6 committed 243 days ago

Verified a21e5b9d

Support complicated use cases with TiedLayerSpec (#7208)

limjcst committed 244 days ago

Verified 185330cd

HPU accelerator memory mapping is broken because of torch fill uninit memory (#7209)

oelayan7 committed 245 days ago

Verified 56005d2b

update dependencies version info (#7206)

inkcherry committed 245 days ago

Verified 29fa95a8

Update to fix pydantic warning (#7193)

loadams committed 250 days ago

Verified 027ee21a

Reland perf fix for nan inf check (#7184)

nelyahu committed 251 days ago

Verified 3c1817f3

Update to new torch grad hook API: BF16Optimizer and Stage2 (#7189)

deepcharm committed 252 days ago

Verified 79ff1627

Add qwen3 autotp support (#7187)

Yejing-Lai committed 252 days ago

Verified 11aca297

Fix issue #5242 grad_norm and loss is nan (#7171)

Glaceon-Hyy committed 255 days ago

Verified 1f706621

async tp allreduce (#7115)

inkcherry committed 256 days ago

Verified b8cc1eb0

Cross layer overlapping for domino (#7178)

hwchen2017 committed 256 days ago

Verified f355b9ea

Update version.txt after 0.16.5 release (#7180)

loadams committed 257 days ago

Verified 3054b934

Variable batch size and LR scheduler (#7104)

bm-synth committed 257 days ago

Verified 20f988ea

Fix typos in GDS blog (#7177)

loadams committed 258 days ago

Verified ac295aa0

fix `seq_parallel_communication_data_type` constant. (#7175)

stas00 committed 258 days ago

Verified 2dbb5d94

Link AutoTP blog in the front page (#7167)

hwchen2017 committed 259 days ago

Verified e7e20e4a

[NFC] Typo fix in SP layer. (#7152)

c8ef committed 260 days ago

Verified 2b245a99

Add destroy to tests to free memory (#7160)

tohtana committed 260 days ago

Verified 9ae010e6

Avoid graph break by removing redundant requires_grad attr change (#7158)

deepcharm committed 260 days ago

Verified d40cf466

hf tp+zero training doc. (#7151)

inkcherry committed 264 days ago

Verified 1ca83a6b

fix leak of z3 buffer

Masahiro Tanaka committed 264 days ago

2e7f8e58

Update container version that runs on A6000 tests. (#7153)

loadams committed 265 days ago

Verified 3388f833

Enhance Gaudi2 CI/Nightly Coverage with Model Parallelism and Linear Tests (#7146)

raza-sikander committed 266 days ago

Verified 29e9fd53

Older