Commits deepspeedai/DeepSpeed

[checkpoint] Execute affine transfer plans with bounded peer staging (#8809)

0z5a committed 5 days ago

Verified f3c6cbb2

ZeRO-3: build the partial-offload device string from the public accelerator API (#8994)

GuCarpenter committed 5 days ago

Verified d9225000

Keep float32 intermediates in the clamped SwiGLU eager fallback (#8992)

GuCarpenter committed 5 days ago

Verified 961df84b

Add an opt-in deterministic execution policy (#8917)

ziming-zh committed 5 days ago

Verified dfc63adf

Update version.txt post-v0.19.9 release (#8999)

loadams committed 5 days ago

Verified 2cd4984a

ZeRO-1/2: preserve gradients across request boundaries until step (#8949)

stas00 committed 6 days ago

Verified dc046b15

[VocabTP] vocab_parallel_ce support fused LM-head projection + loss for the HF forward (#8945)

jinyouzhi committed 6 days ago

Verified 1a956270

Cap CPU accelerator memory at the cgroup limit of the process (#8930)

prabhkodes committed 7 days ago

Verified 2d80cedd

[Hybrid Engine] Enable graph capture for HybridEngine on non-CUDA accelerators (#8933)

VenusTZZ committed 7 days ago

Verified ab7fd98f

ci: pin torch<2.15 for MPS workflow until pytorch/pytorch#202529 is fixed (#8955)

delock committed 8 days ago

Verified ba0d63ce

ZeRO-3: copy offloaded master weights back to the device through pinned buffer (#8840)

pengdurice committed 9 days ago

Verified c6291b7e

Add klxpu (KUNLUNXIN XPU) self-hosted CI for the v1 unit suite (#8923)

GuCarpenter committed 10 days ago

Verified 27360990

[AffineIR] Storage maps of AutoEP experts under EDP and ZeRO (#8941)

jinyouzhi committed 10 days ago

Verified 66fef47b

Sum GPTBigCode's shared key/value gradient across tensor-parallel ranks (#8928)

Achyuthan-S committed 10 days ago

Verified 9148ab22

Check where column-split pieces write in a shard (#8922)

Achyuthan-S committed 10 days ago

Verified 8d4f29c7

Fix race between folded checkpoint load and metadata drop in AutoEP+AutoTP test (#8943)

jinyouzhi committed 10 days ago

Verified 1bdaaf73

[AutoEP] Ensure consistent order of partition groups in all-gather when ZeRO-3 (#8876)

jinyouzhi committed 11 days ago

Verified 12ffb990

Add a v1 multi-rank CPU suite (manual dispatch first) (#8926)

delock committed 11 days ago

Verified 8f9db2c6

[AffineIR] Affine storage maps for live ZeRO parameters and optimizer states (#8885)

jinyouzhi committed 11 days ago

Verified f8185049

Use nvtx's module-level API so NVTX domain ranges work on older nvtx releases (#8900)

hwchen2017 committed 12 days ago

Verified 3dd1293f

Fix performance regression caused by comms logger and bug in comms_logger.enabled (#8909)

hwchen2017 committed 12 days ago

Verified 322283b2

Skip accelerator sync for unprofiled comm ops (#8795)

guptaishaan committed 12 days ago

Verified b5829596

Keep fp32 masters on ZeRO-1 and ZeRO-2 when a resume skips optimizer state (#8891)

sfc-gh-truwase committed 12 days ago

Verified 2dd22dc1

AutoEP: fused Triton kernel for the clamped expert activations (#8652)

pengdurice committed 13 days ago

Verified 7501610e

Bump AWS capacity-fallback Python from 3.10.13 to 3.12.14 (#8921)

delock committed 13 days ago

Verified 8ae7921d

Strip ANSI color codes before parsing FAILED lines in cpu-multi-triage (#8882)

delock committed 13 days ago

Verified 507602ff

Honor fp16 hysteresis in the ZeRO stage 0 fp16 optimizers (#8647)

vineethsaivs committed 13 days ago

Verified 62656fe7

Fix scalar parameter byte-view broadcasts (#8642)

adenzhou1350 committed 13 days ago

Verified bf222a2a

Set TORCHINDUCTOR_COMPILE_THREADS=1 in modal CI sandbox to stop 300s test timeouts (#8914)

delock committed 13 days ago

Verified 1fcac823

[AffineIR] Plan expert transfers from the live runtime placement (#8800)

jinyouzhi committed 14 days ago

Verified 8a47f989