huggingface/accelerate

Pull Requests Commits

Release: v1.13.0

SunMarc committed 67 days ago

e6ee1337

Fix testing ci (#3955)

SunMarc committed 67 days ago

Verified 2a7e27f7

Faster import (#3953)

SunMarc committed 67 days ago

Verified 0990ded5

fix-stateful-dataloader (#3952)

SunMarc committed 67 days ago

Verified 5cf9cf88

Prepare TP fix (#3945)

michaelbenayoun committed 68 days ago

Verified beed693e

Fix StatefulDataLoader KeyError with num_workers > 0 (#3931)

veeceey committed 69 days ago

Verified 8067abae

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

Mr-Neutr0n committed 69 days ago

Verified 8ec83c8a

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

jashshah999 committed 69 days ago

Verified 7554afbc

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

yuxinyuan committed 69 days ago

Verified 23f2ab39

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

ManasVardhan committed 69 days ago

Verified 58c3605f

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

thecaptain789 committed 69 days ago

Verified 2d2b440b

fixing ci (#3948)

SunMarc committed 72 days ago

Verified edd98518

lazy compile disable (#3947)

SunMarc committed 72 days ago

Verified 7ed69be0

Fix cpu offload (#3946)

SunMarc committed 72 days ago

Verified 4721214f

Neuron integration (#3935)

michaelbenayoun committed 73 days ago

Verified 3bffed54

fix: load the HPU RNG state (#3937)

michaelbenayoun committed 74 days ago

Verified ae36c744

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

michaelbenayoun committed 74 days ago

Verified ddfcb63e

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

quic-meetkuma committed 81 days ago

Verified 46ba4818

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

quic-meetkuma committed 83 days ago

Verified 1e1ffd41

fix parallelism_config None error (#3927)

jp1924 committed 86 days ago

Verified 8b4c216b

Remove unused checkpointing_steps argument from parser (#3920)

3manifold committed 93 days ago

Verified 445ef713

DeepSpeed has its own process group (#3916)

kashif committed 93 days ago

Verified e2dde279

Enable evaluation during deepspeed Sequence Parallel (#3917)

jp1924 committed 95 days ago

Verified abb643d0

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

timonmerk committed 103 days ago

Verified b7f2212e

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

hexfaker committed 103 days ago

Verified c231e513

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

yao-matrix committed 104 days ago

Verified 00e4d264

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

kaixuanliu committed 104 days ago

Verified 363816a8

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

kashif committed 104 days ago

Verified 08e212a3

fix tp only bug (#3908)

sywangyi committed 116 days ago

Verified 38dadd95

SunMarc committed 117 days ago

36479b88

Older