Commits huggingface/accelerate

Release: v1.15.0

SunMarc committed 23 days ago

6afc1e5e

Fix nightly CI: tensor parallel size detection and DeepSpeed warmup steps (#4229)

SunMarc committed 23 days ago

Verified 636a9343

docs: fix three dead links left by the docs restructure (#4230)

simpleqt committed 23 days ago

Verified 385d9fb4

docs: fix garbled sentence in CONTRIBUTING.md (#4231)

simpleqt committed 23 days ago

Verified c424d0b8

Clip grad norm support for dtensors (#4219)

michaelbenayoun committed 24 days ago

Verified 0f7e35fe

Fix FSDP2/ PEFT/`FULL_STATE_DICT` dropping every rank's adapter shard except rank 0 (#4206)

AmineDiro committed 24 days ago

Verified b2ac5095

Fix load_accelerator_state only restoring one RNG backend (#4217)

Rakshit-gen committed 24 days ago

Verified 33e8bc49

Treat MPS out-of-memory errors as OOM in find_executable_batch_size (#4227)

caiotheodoro committed 24 days ago

Verified 50de3ad4

feat: add the neuron device branch in state (#4218)

michaelbenayoun committed 24 days ago

Verified ab5fe8e5

Fix MLFLOW_NESTED_RUN never being able to turn nesting off (#4223)

yupengtang committed 24 days ago

Verified 9baf95be

Fix accelerate launch --cpu never setting the KMP variables (#4222)

yupengtang committed 24 days ago

Verified d8544ddf

Fix `estimate-memory` for timm>=1.0.29 by adding the `hf-hub:` prefix (#4213)

iamsharduld committed 24 days ago

Verified 786672cb

Fix accelerate env never reading the MPS chip name (#4221)

yupengtang committed 25 days ago

Verified 9e5d1de5

docs: fix mixed precision argument in examples (#4192)

noobyalan committed 25 days ago

Verified ede1c22b

Fix compile_regions running the uncompiled module (#4188)

hjinnkim committed 25 days ago

Verified a4d394de

Forward reduce_batch_size_fn through the find_executable_batch_size decorator form (#4160)

vineethsaivs committed 25 days ago

Verified e07f2555

docs: fix documented parameter names that do not match the signature (#4156)

massimiliano1991 committed 25 days ago

Verified 29eb9954

fix: handle FP8 dtypes in disk offload write/read path (#4151)

shoemoney committed 25 days ago

Verified 7b16bf3f

Fix memory accounting for integrated CUDA devices (#4187)

harshitakrishna18 committed 25 days ago

Verified 069b3d6d

Reach through the DeepSpeed optimizer wrapper in AcceleratedOptimizer.eval() (#4130)

vineethsaivs committed 25 days ago

Verified e965f100

Handle missing accelerate executable in env command (#4168)

tandede committed 25 days ago

Verified 6bd00de6

chore(deps): bump the actions group across 1 directory with 4 updates (#4182)

dependabot[bot] committed 25 days ago

Verified f627c45e

Report MPS as the accelerator in accelerate env. The accelerator detection chain in (#4158)

xquantize committed 25 days ago

Verified 004717a5

Add type hints to utils/memory.py (#4152)

RudrenduPaul committed 25 days ago

Verified 69db6aa8

Add missing type hints in checkpointing.py (#4153)

RudrenduPaul committed 25 days ago

Verified 8eb8f115

Fix distributedType -> distributed_type typo in downcast_bf16 guard (#4149)

uttam12331 committed 25 days ago

Verified 5d0176a9

Raise a clear error when FSDP is enabled on a mesh with no shard dimension (#4180)

qgallouedec committed 29 days ago

Verified fd132822

prepare_model: don't move DTensor-sharded models to the device (#4181)

qgallouedec committed 29 days ago

Verified f7cc8e9a

Add FSDP2 `activation_checkpointing_offload`: offload checkpointed layer inputs to pinned CPU memory (#4175)

qgallouedec committed 31 days ago

Verified 2ab9b2cc

Refuse context parallelism for models with sliding-window or chunked attention layers (#4177)

qgallouedec committed 32 days ago

Verified bd204169