huggingface/accelerate

Pull Requests Commits

Release: v0.34.2

muellerzr committed 1 year ago

c61f41c6

Release: v0.34.1

muellerzr committed 1 year ago

beb43781

Allow DataLoaderAdapter subclasses to be pickled by implementing `__reduce__` (#3074)

byi8220 committed 1 year ago

e13bef2c

Fix FSDP auto_wrap using characters instead of full str for layers (#3075)

muellerzr committed 1 year ago

73a1531e

Release: v0.34.0

muellerzr committed 1 year ago

159c0dd0

Remove `skip_first_batches` support for StatefulDataloader and fix all the tests (#3068)

muellerzr committed 1 year ago

Verified 8931e5e4

Speed up tests by shaving off subprocess when not needed (#3042)

muellerzr committed 1 year ago

Verified a8485924

add set_epoch for MpDeviceLoaderWrapper (#3053)

append-only committed 1 year ago

Verified 758d6243

Fix typo in comment (#3045)

mokizzz committed 1 year ago

Verified b07ad2ad

use duck-typing to ensure underlying optimizer supports schedulefree hooks (#3055)

tmm1 committed 1 year ago

Verified 1d09a20f

Do not import `transformer_engine` on import (#3056)

oraluben committed 1 year ago

Verified 3fcc9461

Update torchpippy (#2938)

muellerzr committed 1 year ago

Verified 939ce400

Add FP8 docker images (#3048)

muellerzr committed 1 year ago

Verified c2120927

Add a SLURM example with minimal config (#2950)

muellerzr committed 1 year ago

Verified 654e1d99

Update CONTRIBUTING.md Setup Instructions (#3046)

siddk committed 1 year ago

Verified 8c3aded2

Decouple `prepare_data_loader()` from Accelerator (#3047)

siddk committed 1 year ago

Verified 27899339

Fixup dataloader state dict bugs + incorporate load/save_state API (#3034)

muellerzr committed 1 year ago

Verified 726140ca

Fix batch_sampler maybe None error (#3025)

candlewill committed 1 year ago

Verified 2d4f1dda

Fix fp8 benchmark on single GPU (#3032)

muellerzr committed 1 year ago

Verified c0cf860d

Add early support for `torchdata.stateful_dataloader.StatefulDataLoader` within the `Accelerator` (#2895)

byi8220 committed 1 year ago

Verified ad3f574a

Improve config handling and add a zoo (#3029)

muellerzr committed 1 year ago

Verified 1a6af0bd

Add end_training/destroy_pg to everything and unpin numpy (#3030)

muellerzr committed 1 year ago

Verified 52fae096

Fix torch version check (#3024)

muellerzr committed 1 year ago

Verified 7ffe7662

Set correct NPU backend and distributed_type when using transfer_to_npu (#3021)

ArthurinRUC committed 1 year ago

Verified 5536a3a8

Tweak defaults for quantized-typed FP8 TE weights (#3018)

muellerzr committed 1 year ago

Verified 7ec8eab9

destroy process group in `end_training` (#3012)

SunMarc committed 1 year ago

Verified 589fddd3

Wrong import check for TE (#3016)

muellerzr committed 1 year ago

Verified 99c69aaf

fix default value for rank size in cpu threads_per_process assignment logic (#3009)

rbrugaro committed 1 year ago

Verified 00785cd9

Enable FSDP & Deepspeed + FP8 (#2983)

muellerzr committed 1 year ago

Verified a452327e

Fix `find_tied_params` for models with shared layers (#2986)

qubvel committed 1 year ago

Verified 851cf343

Older