microsoft/DeepSpeed

Pull Requests Commits

Fix OnebitLamb NaN propagation with empty parameters (#7736)

Rakshit-gen committed 39 minutes ago

Verified b4e74a91

Fix DecoupledCheckpointEngine deadlock and improve reliability (#7742)

Rakshit-gen committed 2 days ago

Verified ba29fdca

Fix Nebula checkpoint engine commit() API mismatch (#7740)

Rakshit-gen committed 2 days ago

Verified 6d9c3dc0

Add core api update blog (#7738)

tohtana committed 4 days ago

Verified 20cfce00

[BUG] Fix UlyssesSPAttentionHF.register_with_transformers() crash with PEFT models (#7737)

Rakshit-gen committed 5 days ago

Verified b6f4e20e

replace moe checkpoint dp_world_size with seq_dp_world_size (#7732)

wukong1992 committed 5 days ago

Verified 377a0d15

Fix #7733: Replace torch.sqrt with math.sqrt in scale_lr for sqrt method (#7735)

Rakshit-gen committed 5 days ago

Verified d7d4eebc

Fix rare hang in DeepSpeed Async I/O wait by releasing the Python GIL (#7727)

xylian86 committed 6 days ago

Verified 2bc16e21

Fix testcases that depends on triton (#7731)

k-artem committed 7 days ago

Verified 5373a880

[Engine] Only scale gradients if scale_wrt_gas is True (#7724)

kashif committed 12 days ago

Verified d568375e

Skip none in backward hook (#7725)

tohtana committed 12 days ago

Verified fffcf2f5

fix typo (#7722)

stas00 committed 12 days ago

Verified 49b3d62b

[AMD][ROCm] Improve support of AMD (#7448)

k-artem committed 14 days ago

Verified b00b75f0

Fix SuperOffloadOptimizer_Stage3 crash due to missing param_names parameter (#7715)

ImaGoodFella committed 14 days ago

Verified 0e77020c

Disable deterministic option in compile tests (#7720)

tohtana committed 14 days ago

Verified 4862115b

Update version (#7719)

sfc-gh-truwase committed 15 days ago

Verified 2fda5cec

Wall clock timers API (#7714)

sfc-gh-truwase committed 15 days ago

Verified 6eb98aaf

Enabled compiled autograd for backward pass (#7667)

deepcharm committed 19 days ago

Verified e0a6bb51

Low-precision master params/grads/optimizer states (#7700)

tohtana committed 20 days ago

Verified 39a682d7

fix: skip aio wait when swap tensors is empty (#7712)

xylian86 committed 20 days ago

Verified 78edd6f7

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

zhengchenyu committed 20 days ago

Verified c069ceb3

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

rraminen committed 21 days ago

Verified 28fbb808

Put Muon optimizer momentum buffer on GPU (#7648)

delock committed 28 days ago

Verified 7f2f4232

Add news about Ray x DeepSpeed Meetup (#7704)

PKUWZP committed 30 days ago

Verified 3c8c6f34

PyTorch-compatible backward API (#7665)

tohtana committed 35 days ago

Verified 53e91a09

Trust intel server for XPU tests (#7698)

tohtana committed 35 days ago

Verified 51dc8884

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

delock committed 36 days ago

Verified a83fd7b4

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

loadams committed 36 days ago

Verified 5b05fb5b

Use pytorch utils to detect ninja (#7687)

Emrys-Merlin committed 41 days ago

Verified 4cb26cab

modal ci: fix group concurrency (#7691)

stas00 committed 41 days ago

Verified dff81cb6

Older