huggingface/accelerate

Pull Requests Commits

unpin datasets (#3681)

SunMarc committed 1 year ago

Verified 6e104f31

Speedup model loading by 4-5x in Diffusers ⚡ (#3674)

a-r-r-o-w committed 1 year ago

Verified 524e5f98

Bunch of FSDP improvements (#3671)

S1ro1 committed 1 year ago

Verified d6c986c3

xpu enablement on left cases (#3654)

yao-matrix committed 1 year ago

Verified 1ac8643d

Fix: properly error when DDP + Dtensor model (#3629)

S1ro1 committed 1 year ago

Verified 07ce7486

Added a check in the no_sync() function to avoid errors when using deepspeed zero2/3. (#3656)

xliu0105 committed 1 year ago

Verified 175fe915

Fix fsdp2 example (#3657)

shimizust committed 1 year ago

Verified fe16ce8b

Update gradient_accumulation.md (#3649)

kilavvy committed 1 year ago

Verified 5987d79a

SunMarc committed 1 year ago

Verified 31af8d4e

Add support for e5e2 and default to hybrid when launcher is used (#3640)

IlyasMoutawwakil committed 1 year ago

Verified b7493a82

bump to v1.9.0dev

SunMarc committed 1 year ago

a16d2bb3

fix grad acc deepspeed (#3638)

SunMarc committed 1 year ago

Verified cac22ed9

Fix: correct labels (#3637)

S1ro1 committed 1 year ago

Verified be826a6b

Feat: add cpu offload (#3636)

S1ro1 committed 1 year ago

Verified 59396408

[DeepSpeed] sync gradient accum steps from deepspeed plugin (#3632)

kashif committed 1 year ago

Verified 7f9c8cbe

feat: use datasets.IterableDataset shard if possible (#3635)

SunMarc committed 1 year ago

Verified 9888c7ed

Fix Typos in Documentation and Comments (#3621)

leopardracer committed 1 year ago

Verified 42a68c30

Integrate SwanLab for offline/online experiment tracking for Accelerate (#3605)

ShaohonChen committed 1 year ago

Verified 6597dae7

remove hardcoded cuda from fsdpv2 (#3631)

IlyasMoutawwakil committed 1 year ago

Verified 8878d937

remove ipex.optimize in accelerate (#3608)

yao-matrix committed 1 year ago

Verified 2eaf5cdb

[Deepspeed] deepspeed auto grad accum (#3630)

kashif committed 1 year ago

Verified 23c1d8db

add xpu support in TorchTensorParallelPlugin (#3627)

yao-matrix committed 1 year ago

Verified 0af621bb

Add fp8_e5m2 support in `dtype_byte_size` (#3625)

SunMarc committed 1 year ago

Verified bee04f1b

fix xpu 8bit value loading (#3623)

jiqing-feng committed 1 year ago

Verified 8a953f08

small fix (#3619)

IlyasMoutawwakil committed 1 year ago

Verified 3518c035

Remove device_count (#3587)

sorgfresser committed 1 year ago

Verified 2f8fd72e

[FSDP2] Refactor + FP8 (#3585)

S1ro1 committed 1 year ago

Verified d2e6b031

better handle FP8 with and without deepspeed (#3611)

IlyasMoutawwakil committed 1 year ago

Verified b9fee48c

Fix bf16 training with TP (#3610)

SunMarc committed 1 year ago

Verified 3a82b056

fix deepspeed regional compilation (#3609)

IlyasMoutawwakil committed 1 year ago

Verified 6b61a373

Newer Older