Pull Requests huggingface/accelerate

Disable hook compile

#3888 by SunMarc was merged 2025-12-17 15:53

Update support of Megatron-LM PR 2

#3887 by pengdurice was merged 2025-12-16 13:00

Fix: Remove duplicate W&B initialization in offline mode (#3818)

#3886 by shantanugupta2004 was merged 2025-12-16 13:10

using `spawn` instead of `fork` for XPU device

#3884 by kaixuanliu was merged 2025-12-15 11:22

Remove ipex

#3883 by yao-matrix was merged 2025-12-15 11:22

Fix KeyError in extract_model_from_parallel for partial torch.compile

#3881 by amanzoni1 was merged 2025-12-16 13:00

[DeepSpeed] scale grad for zero-2

#3880 by kashif was closed 2025-12-11 20:37

Fix FSDP2 tied embedding errors with targeted ValueError guidance

#3878 by amanzoni1 was merged 2025-12-11 13:02

Avoid using nvidia-smi on a CPU-only Colab instance

#3872 by FlorianVal was merged 2025-12-04 15:35

fix: mixed_precision param from accelerate config should be used for FSDP mp

#3864 by kmehant was closed 2025-12-02 09:03

[SP and CP] error out if both CP and SP enabled

#3862 by kashif was merged 2025-11-28 14:36

[SP] fix loss computation example bug

#3858 by kashif was merged 2025-11-28 10:43

add MS-AMP deprecation warnings

#3857 by neha222222 was merged 2025-12-08 13:41

Fix execution with Transformer Engine

#3852 by ksivaman was merged 2025-12-01 13:36

Update PR template, setup.py author email

#3851 by tomaarsen was merged 2025-11-25 11:15

Allow non-Tensor values in a batch with `dispatch_batches=True`

#3850 by tomaarsen was merged 2025-11-26 16:57

Upcast FSDP2 parameters only if requires_grad

#3848 by ojh31 was merged 2025-11-26 15:03

feat: added fine tuning example focused on TPUs

#3847 by tengomucho was merged 2025-11-24 16:40

fix module and optimizer parameter mismatch before prepare_tp_

#3845 by naomili0924 was merged 2025-11-27 11:00

device type helper

#3843 by kashif was merged 2025-11-21 11:11

Updating support of Megatron-LM

#3842 by pengdurice was merged 2025-12-03 13:15

use self hosted runner

#3841 by SunMarc was merged 2025-11-20 13:12

Add num_processes and parallelism_config parameters

#3837 by MihaiB-dev was closed 2025-12-21 15:07

[Bug] Update torch.optim.Optimizer parameter states after tensor parallelism

#3835 by naomili0924 was merged 2025-11-19 13:04

ArXiv -> HF Papers

#3834 by qgallouedec was merged 2025-11-10 12:51

Fix FP8 torchao default config with padding and FSDP2 all-gather support

#3831 by shimizust was merged 2025-12-03 13:16

update typo in bnb quantisation 4bit flag docstring

#3828 by hbraith was merged 2025-11-04 14:26

Fix typo in broadcast_object_list docstring

#3823 by wsntxxn was merged 2025-11-13 15:01

decouple backward and step in accelerator/deepspeed

#3819 by naomili0924 was closed 2025-12-16 15:09

Deepspeed Ulysses/ALST integration

#3817 by stas00 was merged 2025-11-20 17:24