microsoft/DeepSpeed

Pull Requests Commits

Raise Modal sandbox timeout to absorb full-suite runtime

delock committed 2 days ago

f9aad0d7

docs: stop Codex false-alarm on DCO sign-off (#8561)

delock committed 3 days ago

Verified 327190be

Deprecate sparse_gradients (#8504)

pengdurice committed 3 days ago

Verified 5166f494

Deprecate loco zero++ (#8534)

hwchen2017 committed 3 days ago

Verified c8079715

Update version.txt after 0.19.7 release (#8555)

loadams committed 3 days ago

Verified ebac744c

Deprecate leftovers of compression library (#8535)

hwchen2017 committed 3 days ago

Verified 8ae0509e

[bugfix] _DimZeroAllToAll silently fails to send grads under torch.compile (#8491)

stas00 committed 3 days ago

Verified f5af15c2

Cache the Modal sandbox install chain in image layers (#8536)

delock committed 3 days ago

Verified 83dd543f

Emit affine maps from AutoTP layers (#8519)

Achyuthan-S committed 3 days ago

Verified 69797b0b

Add DCO sign-off to release commits (#8527)

loadams committed 4 days ago

Verified 7075430a

Update version.txt after 0.19.6 release (#8333)

loadams committed 4 days ago

Verified 78db78d8

Give Muon's momentum the dtype of the gradient it is combined with (#8483)

alanhuangyoo committed 4 days ago

Verified 431f8ace

Read rope_theta from rope_parameters across Inference V2 (#8345)

alanhuangyoo committed 5 days ago

Verified 1b892a75

Partition AutoEP expert parameters per layer under ZeRO-3 (#8424)

tohtana committed 5 days ago

Verified 6d5c56f3

Gate the offload-state memory deltas on allocator-backed stats (#8409)

delock committed 5 days ago

Verified e6d2d409

Probe the device module for train_cifar's fork_rng device entries (#8407)

delock committed 5 days ago

Verified 03063002

feat(rollout): add continuous batching profiling (#8494)

nathon-lee committed 5 days ago

Verified 8950dc33

Deprecate sparse attention (#8493)

hwchen2017 committed 5 days ago

Verified a3f9d4aa

Make wait() idempotent on AllGatherHandle and NoGatherHandle (#8487)

pengdurice committed 5 days ago

Verified 10e358b8

Honor --include/--exclude in the SLURM launcher (#8304)

vineethsaivs committed 5 days ago

Verified 8bdd8f19

Offload a saved view when it is the last value holding its storage (#8388)

pengdurice committed 5 days ago

Verified 56631282

Fix sequence overlap backward gradient permutation (#8342)

taking-lying-flat committed 5 days ago

Verified 2ce20d6a

Validate positive inference and HybridEngine max output tokens (#8343)

taking-lying-flat committed 5 days ago

Verified 7c1b0f2c

[Phase 2] Add NEON SIMD path for CPU Adam on AArch64 (#8453)

PKUWZP committed 6 days ago

Verified c102bbd1

Muon silently discards the param groups it is given (#8440)

alanhuangyoo committed 6 days ago

Verified 113c6752

Remove triton compatibility check for fp_quantizer (#8492)

hwchen2017 committed 6 days ago

Verified cc05b811

[muon] Keep the momentum out of steps the loss scaler discards (#8435)

alanhuangyoo committed 7 days ago

Verified b5e000c4

Deprecate unused DeepSpeed features (#8490)

sfc-gh-truwase committed 8 days ago

Verified 71d316d6

[muon] Reconcile the momentum dtype when a checkpoint is restored (#8433)

alanhuangyoo committed 8 days ago

Verified b726f4ed

Carry the affine scale on the replicated map, not the split (#8477)

Achyuthan-S committed 8 days ago

Verified bce0adf0

Older