Commits huggingface/accelerate

chore(deps): bump the actions group with 4 updates

dependabot[bot] committed 3 days ago

Verified fdb9f33f

Ask transformers whether a model supports context parallelism (#4198)

qgallouedec committed 3 days ago

Verified ccfdff7d

Skip `_no_split_modules` classes the model doesn't contain when auto-wrapping (#4384)

qgallouedec committed 6 days ago

Verified 4ebc7268

Allow tensor parallelism next to replicated data parallelism (#4193)

qgallouedec committed 8 days ago

Verified 419d0e0c

FSDP2: move placeholder weights to meta before the fp32 upcast (#4369)

pagel-s committed 9 days ago

Verified 2f4c053a

Bump doc-builder workflow pin (#4368)

paulinebm committed 10 days ago

Verified 08198562

chore(deps): bump the actions group with 4 updates (#4366)

dependabot[bot] committed 10 days ago

Verified 7d8824f1

Fix metric padding removal after skip_first_batches (#4235)

Excelius-Wang committed 14 days ago

Verified 01c73fbd

Fix wait_for_everyone() crashing on MULTI_CPU when the machine has an accelerator (#4354)

JoeyTan21 committed 14 days ago

Verified a68e0d50

[docs] Fix 31 dead or moved links across seven guides (#4348)

pratikgx committed 15 days ago

Verified aebbcc7b

Fix loading checkpoints with modules offloaded to disk (#4352)

yupengtang committed 16 days ago

Verified ea1311b1

Fix IndexError when a tied parameter's partner module is split (#4344)

simpleqt committed 16 days ago

Verified a3b865e6

Fix notebook_launcher to propagate mixed_precision to the workers (#4331)

simpleqt committed 16 days ago

Verified d737ded7

Fix skip_first_batches dropping drop_last, non_blocking, slice_fn, and mesh (#4296)

truecallerabreham committed 16 days ago

Verified a7c67eca

chore(deps): bump the actions group with 7 updates (#4350)

dependabot[bot] committed 16 days ago

Verified 313d4839

Fix accelerator dispatch test on minimum PyTorch (#4336)

SunMarc committed 20 days ago

Verified 421af8be

Fix unwrap_model(keep_fp32_wrapper=False) breaking hooked models (#4255)

yupengtang committed 20 days ago

Verified d3c81857

Fix ParallelismConfig.to_json returning None instead of the config dict (#4326)

Linxiushen committed 20 days ago

Verified 0bddf0eb

Prefer all_gather_single over deprecated all_gather_into_tensor (#4272)

Yi-111-a committed 20 days ago

Verified d8dde8fb

Fix split_between_processes on tuples with apply_padding=True (#4307)

shoutoutuoadi325 committed 20 days ago

Verified 37569f11

Fix `balanced_low_0` reading max_memory by position instead of device id (#4298)

Om-singhaI committed 20 days ago

Verified 47bd9b98

Fix dtype_byte_size reporting zero bytes for sub-byte and packed dtypes (#4237)

VihaanAgarwal committed 20 days ago

Verified 8afee62c

Reduce key merging overhead in OffloadedWeightsLoader (#4281)

emme1t committed 20 days ago

Verified 65f83e49

Do not import torch.distributed in model_has_dtensor on a non-distributed build (#4250)

shoemoney committed 20 days ago

Verified 05bf97ef

Stop accelerate launch from forcing torch.compile dynamic=False (#4254)

yupengtang committed 20 days ago

Verified af223223

Fix custom accelerator support (#4335)

SunMarc committed 20 days ago

Verified 03231349

Read ACCELERATE_BYPASS_DEVICE_MAP case-insensitively, like every other flag (#4241)

Sreekant13 committed 20 days ago

Verified 51081f7b

chore(deps): bump the actions group with 3 updates (#4299)

dependabot[bot] committed 20 days ago

Verified 38fc23de

Fix VRAM leak from cached int8 weights during CPU/disk offload (#4325)

edmcman committed 20 days ago

Verified 38caef13

Fix FSDP2 fp16 checkpointing advancing the GradScaler it is restoring (#4291)

HaomingSong committed 21 days ago

Verified fb850911