huggingface/accelerate

Pull Requests Commits

test extract parallel model

IlyasMoutawwakil committed 1 year ago

b65a26f0

default is True

IlyasMoutawwakil committed 1 year ago

23f95bb7

safe inplace regional compilation as a deafult

IlyasMoutawwakil committed 1 year ago

bcc3a2ec

revert repeated class check

IlyasMoutawwakil committed 1 year ago

f351bbd1

IlyasMoutawwakil committed 1 year ago

325e80ce

IlyasMoutawwakil committed 1 year ago

9d7efcaa

IlyasMoutawwakil committed 1 year ago

1eb4a956

IlyasMoutawwakil committed 1 year ago

c7388024

IlyasMoutawwakil committed 1 year ago

de3dfa15

remove unecessary pops

IlyasMoutawwakil committed 1 year ago

ac0e5658

make compile regions inplace to avoid recursion in some scenarios

IlyasMoutawwakil committed 1 year ago

386d874b

IlyasMoutawwakil committed 1 year ago

53f33be4

IlyasMoutawwakil committed 1 year ago

39bdbc80

Fix: Defer Tracker Initialization to Prevent Premature Distributed Setup (#3581)

yuanjua committed 1 year ago

Verified db536cbf

enable regional_compilation benchmark on xpu (#3592)

yao-matrix committed 1 year ago

Verified 4e9d0deb

Add kwargs to optimizer, scheduler and dataloader using function `accelerator().load_state()` (#3540)

luiz0992 committed 1 year ago

Verified 8cb3ace8

Resolve logger warnings (#3582)

emmanuel-ferdman committed 1 year ago

Verified b6d97cb8

Add support for standalone mode when default port is occupied on single node (#3576)

laitifranz committed 1 year ago

Verified 33967d47

enable test_cli & test_example cases on XPU (#3578)

yao-matrix committed 1 year ago

Verified 5b1fcda3

goodbye torch_ccl (#3580)

yao-matrix committed 1 year ago

Verified f55f0533

enable test_load_checkpoint_and_dispatch_with_broadcast cases on XPU (#3579)

yao-matrix committed 1 year ago

Verified 1ec99f0b

bump to v1.8.0dev

SunMarc committed 1 year ago

417bc529

enable test_dispatch_model_tied_weights_memory_with_nested_offload_cpu on xpu (#3569)

yao-matrix committed 1 year ago

Verified 97c93c48

set backend correctly for CUDA+FSDP2+cpu-offload (#3574)

SunMarc committed 1 year ago

Verified cd37bbb6

Fix prevent duplicate GPU usage in distributed processing (#3526)

ved1beta committed 1 year ago

Verified 7aa3b56c

reenable FSDP2+qlora support (#3546)

winglian committed 1 year ago

Verified 14f4306c

Add regional compilation to cli tools and env vars (#3572)

IlyasMoutawwakil committed 1 year ago

Verified e6e71758

tune env command output (#3570)

yao-matrix committed 1 year ago

Verified 1f6efcea

simplify model.to logic (#3562)

yao-matrix committed 1 year ago

Verified 9fa97f96

add xpu synchronize (#3563)

faaany committed 1 year ago

Verified 764eee4a

Older