huggingface/accelerate

Pull Requests Commits

Faster import (#3953)

SunMarc committed 91 days ago

Verified 0990ded5

fix-stateful-dataloader (#3952)

SunMarc committed 91 days ago

Verified 5cf9cf88

Prepare TP fix (#3945)

michaelbenayoun committed 92 days ago

Verified beed693e

Fix StatefulDataLoader KeyError with num_workers > 0 (#3931)

veeceey committed 93 days ago

Verified 8067abae

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

Mr-Neutr0n committed 93 days ago

Verified 8ec83c8a

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

jashshah999 committed 93 days ago

Verified 7554afbc

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

yuxinyuan committed 93 days ago

Verified 23f2ab39

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

ManasVardhan committed 93 days ago

Verified 58c3605f

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

thecaptain789 committed 93 days ago

Verified 2d2b440b

fixing ci (#3948)

SunMarc committed 96 days ago

Verified edd98518

lazy compile disable (#3947)

SunMarc committed 96 days ago

Verified 7ed69be0

Fix cpu offload (#3946)

SunMarc committed 96 days ago

Verified 4721214f

Neuron integration (#3935)

michaelbenayoun committed 97 days ago

Verified 3bffed54

fix: load the HPU RNG state (#3937)

michaelbenayoun committed 98 days ago

Verified ae36c744

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

michaelbenayoun committed 98 days ago

Verified ddfcb63e

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

quic-meetkuma committed 105 days ago

Verified 46ba4818

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

quic-meetkuma committed 107 days ago

Verified 1e1ffd41

fix parallelism_config None error (#3927)

jp1924 committed 110 days ago

Verified 8b4c216b

Remove unused checkpointing_steps argument from parser (#3920)

3manifold committed 117 days ago

Verified 445ef713

DeepSpeed has its own process group (#3916)

kashif committed 117 days ago

Verified e2dde279

Enable evaluation during deepspeed Sequence Parallel (#3917)

jp1924 committed 119 days ago

Verified abb643d0

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

timonmerk committed 127 days ago

Verified b7f2212e

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

hexfaker committed 127 days ago

Verified c231e513

enhance new codes to XPU, and make them be device agnostic (#3890)

yao-matrix committed 127 days ago

Verified 00e4d264

fix bug when both cpu_ram_efficient_loading and cpu_offload are enabled (#3910)

kaixuanliu committed 127 days ago

Verified 363816a8

skip device mesh creation when deepspeed and sp_size >1 (#3914)

kashif committed 127 days ago

Verified 08e212a3

fix tp only bug (#3908)

sywangyi committed 140 days ago

Verified 38dadd95

SunMarc committed 141 days ago

36479b88

Remove 8bit force hook for bnb (#3907)

SunMarc committed 141 days ago

Verified 4f8db612

fix crash in optimizer.step when fsdp2 is enabled and model is bfloat16 (#3905)

sywangyi committed 141 days ago

Verified 758f3b18

Newer Older