microsoft/DeepSpeed

Pull Requests Commits

address feedback

stas00 committed 3 days ago

6577c03f

[CI] diff driven test selection

stas00 committed 3 days ago

f84aec4e

Update version post release (#8070)

loadams committed 5 days ago

Verified ad026a1f

Mixed-precision: per-policy param/buffer dtype cast (preserve fp32 buffers) (#8066)

sfc-gh-truwase committed 5 days ago

Verified b919284a

Add Biren SUPA accelerator support (#8054)

frozenleaves committed 6 days ago

Verified 7ad41089

Pass expected grad dtype to register_z3_param in ZeRO-3 release test (#8063)

tohtana committed 8 days ago

Verified 3a917de5

Fix: ZenFlow Adam integration for updated PyTorch backward flow (#7759) (#7771)

Antlera committed 10 days ago

Verified b2145896

Add AutoEP (#7938)

tohtana committed 10 days ago

Verified 7c3fbdc0

Consistent fp32 grads flow (#8056)

sfc-gh-truwase committed 11 days ago

Verified 56e7e0c5

zero3: invalidate coordinator trace on hook re-registration (#8043)

roycho96 committed 11 days ago

Verified 87f31d55

fix(transformer): use correct stride in Transpose_Kernel shared memory indexing to eliminate bank conflicts (#8055)

flutist committed 11 days ago

Verified 65e5f6fc

Remove AutoSP assertion against Transformers version (#8044)

tohtana committed 12 days ago

Verified a14e4e18

Normalize ZeRO-3 DeepCompile grad dtype before reduction (#8038)

tohtana committed 12 days ago

Verified 4f5fb834

Fix DeepCompile ZeRO-1 grad target lifetime (#8036)

tohtana committed 12 days ago

Verified 11eeb7cd

Avoid HF Hub access in CPU unit test setup (#8053)

tohtana committed 13 days ago

Verified 92cfacb9

Add test coverage for Muon muon_lr/adam_lr overrides (#8047)

sowndappan5 committed 16 days ago

Verified 3dc98deb

Fix ZenFlow ZeRO-3 selective optimizer crash with parameter offload on nvme (#8042)

Antlera committed 17 days ago

Verified 28a196f7

Fix DeepCompile ZeRO-3 release parameter lifetime (#8032)

tohtana committed 20 days ago

Verified 3e486feb

Bump version after 0.19.1 release (#8040)

tohtana committed 22 days ago

Verified 334230ff

Version fix to unblock pypi (#8039)

sfc-gh-truwase committed 22 days ago

Verified d5d118c2

Fix DeepCompile all-gather scheduler candidate selection (#8033)

tohtana committed 22 days ago

Verified 819af0e5

Simplify module_inject.transpose (#8028)

xbcReal committed 23 days ago

Verified 1afa3a53

feat(zero): enable torch.func transforms on engine for ZeRO 0/1/2 (#8026)

roycho96 committed 25 days ago

Verified 4370aa50

Add engine.coalesce_grad_reduction() for ZeRO 1/2/3 multi-backward (#7992)

roycho96 committed 25 days ago

Verified 60b242af

Bump version (#8030)

sfc-gh-truwase committed 25 days ago

Verified a0b98d34

Keep required CI checks visible for ignored paths (#8019)

tohtana committed 25 days ago

Verified 77a249f1

Fix full CI test isolation for ZeRO chmod and NVMe quantization tests (#8008)

tohtana committed 25 days ago

Verified 4570c508

Fix PR-target workflow concurrency groups (#8017)

tohtana committed 25 days ago

Verified b7aef4dc

Run AutoSP compile tests sequentially (#8020)

tohtana committed 25 days ago

Verified 2c8a007b

fix(fp16): filter requires_grad in FP16 optimizer flat buffer init (#8029)

avicooper1 committed 25 days ago

Verified 01e3a28b

Older