huggingface/accelerate

Pull Requests Commits

Fix is_hf_initialized attribute (#3976)

SunMarc committed 2 days ago

Verified 13640d32

Torchao offload (#3973)

SunMarc committed 5 days ago

Verified 3475dcfa

fix: conditionally import torch.distributed.algorithms.join in accelerator (#3962)

0xDELUXA committed 8 days ago

Verified 9bbee85f

allow kernels flash-attn (#3959)

kashif committed 9 days ago

Verified f317f824

Fix iterable dataset sharding condition when n_shards == num_processes (#3958)

SunMarc committed 13 days ago

Verified 3e27a7c6

SunMarc committed 14 days ago

74427420

Fix testing ci (#3955)

SunMarc committed 14 days ago

Verified 2a7e27f7

Faster import (#3953)

SunMarc committed 14 days ago

Verified 0990ded5

fix-stateful-dataloader (#3952)

SunMarc committed 15 days ago

Verified 5cf9cf88

Prepare TP fix (#3945)

michaelbenayoun committed 16 days ago

Verified beed693e

Fix StatefulDataLoader KeyError with num_workers > 0 (#3931)

veeceey committed 17 days ago

Verified 8067abae

Fix FSDP2 crash with ignored_params on torch < 2.7.0 (#3924)

Mr-Neutr0n committed 17 days ago

Verified 8ec83c8a

Fix mutable default in Megatron init and IndexError on empty ModuleList (#3944)

jashshah999 committed 17 days ago

Verified 7554afbc

Fix logging logic when in_order is set to True (#3280)

yuxinyuan committed 17 days ago

Verified 23f2ab39

docs: update low-precision training docs to reflect MS-AMP deprecation (#3929)

ManasVardhan committed 17 days ago

Verified 58c3605f

fix: correct typo 'guarentee' to 'guarantee' (#3922)

thecaptain789 committed 17 days ago

Verified 2d2b440b

fixing ci (#3948)

SunMarc committed 19 days ago

Verified edd98518

lazy compile disable (#3947)

SunMarc committed 20 days ago

Verified 7ed69be0

Fix cpu offload (#3946)

SunMarc committed 20 days ago

Verified 4721214f

Neuron integration (#3935)

michaelbenayoun committed 21 days ago

Verified 3bffed54

fix: load the HPU RNG state (#3937)

michaelbenayoun committed 22 days ago

Verified ae36c744

fix: use rng_type for HPU to properly set the RNG state (#3936)

michaelbenayoun committed 22 days ago

Verified ddfcb63e

Revert "Added support for qaic backend for Qualcomm's AI Accelerator (#3926)" (#3928)

quic-meetkuma committed 29 days ago

Verified 46ba4818

Added support for qaic backend for Qualcomm's AI Accelerator (#3926)

quic-meetkuma committed 31 days ago

Verified 1e1ffd41

fix parallelism_config None error (#3927)

jp1924 committed 34 days ago

Verified 8b4c216b

Remove unused checkpointing_steps argument from parser (#3920)

3manifold committed 41 days ago

Verified 445ef713

DeepSpeed has its own process group (#3916)

kashif committed 41 days ago

Verified e2dde279

Enable evaluation during deepspeed Sequence Parallel (#3917)

jp1924 committed 43 days ago

Verified abb643d0

Update command export to use SCRIPT and SCRIPT_ARGS (#3906)

timonmerk committed 51 days ago

Verified b7f2212e

Fix KMP_AFFINITY incorrectly set for non-CPU training (#3912)

hexfaker committed 51 days ago

Verified c231e513

Older