microsoft/DeepSpeed

Pull Requests Commits

Fix that ds_secondary_tensor may be dirty when loading the model or zero checkpoint for zero++. (#7707)

zhengchenyu committed 66 days ago

Verified c069ceb3

[ROCm] Relax tolerances for FP8 unit test for fp16 and bf16 cases (#7655)

rraminen committed 67 days ago

Verified 28fbb808

Put Muon optimizer momentum buffer on GPU (#7648)

delock committed 74 days ago

Verified 7f2f4232

Add news about Ray x DeepSpeed Meetup (#7704)

PKUWZP committed 76 days ago

Verified 3c8c6f34

PyTorch-compatible backward API (#7665)

tohtana committed 81 days ago

Verified 53e91a09

Trust intel server for XPU tests (#7698)

tohtana committed 81 days ago

Verified 51dc8884

Add Qwen2.5 to AutoTP model list (#7696)

delock committed 82 days ago

Verified a83fd7b4

Update SECURITY.md to point to GitHub reporting rather than Microsoft (#7692)

loadams committed 82 days ago

Verified 5b05fb5b

Use pytorch utils to detect ninja (#7687)

Emrys-Merlin committed 87 days ago

Verified 4cb26cab

modal ci: fix group concurrency (#7691)

stas00 committed 87 days ago

Verified dff81cb6

stage 1-2: don't pin memory if not configured (#7689)

stas00 committed 87 days ago

Verified c502d3ac

make debug utils more resilient (#7690)

stas00 committed 87 days ago

Verified 7cb37efc

see_mem_usage: make always work (#7688)

stas00 committed 87 days ago

Verified c2c090b7

allow seperate learning rate "muon_lr" and "adam_lr" for muon optimizer (#7658)

delock committed 89 days ago

Verified df59f203

disable nv-lightning-v100.yml cI (#7681)

stas00 committed 92 days ago

Verified 283f6f5f

leaf modules: explain better (#7674)

stas00 committed 92 days ago

Verified d25fd486

[modal ci] fixes (#7676)

stas00 committed 93 days ago

Verified b073a557

Update version.txt after release (#7675)

loadams committed 94 days ago

Verified c4b8c3f3

README refresh (#7668)

sfc-gh-truwase committed 95 days ago

Verified e993fea3

Fix typo in pytorch-profiler.md documentation (#7652)

kunheek committed 96 days ago

Verified 4691bebd

resolved a 0-dim tensor slicing bug from _get_state_without_padding (#7659)

therealnaveenkamal committed 96 days ago

Verified 6fe3c7ab

UlyssesSP: TiledMLP doc - recomputes forward twice (#7664)

stas00 committed 96 days ago

Verified 76a40757

[bug]: fixed comm_dtype in extra_large_param_to_reduce (#7660)

therealnaveenkamal committed 100 days ago

Verified 3e64f495

Fix misplaced overflow handling return in fused_optimizer.py (#7645)

rraminen committed 100 days ago

Verified d56e847b

ALST/UlyssesSP: more intuitive API wrt variable seqlen (#7656)

stas00 committed 102 days ago

Verified 02da3732

ulysses mpu: additional api (#7649)

stas00 committed 103 days ago

Verified 433e3c72

Deduplicate fp32 weights under torch autocast and ZeRO3 (#7651)

eternalNight committed 103 days ago

Verified 706f6e82

Update version after 0.18.1 release (#7647)

loadams committed 107 days ago

Verified 9a012d21

Update version.txt in advance of next release

loadams committed 107 days ago

3631712b

Add DataStates-LLM: Asynchronous Checkpointing Engine Support (#7166)

mauryaavinash95 committed 108 days ago

Verified d1e62ff2

Newer Older