Minimum required accelerate library is not compatible

danielnelson commented 2 weeks ago

System Info

transformers version: 4.41.2
Platform: Linux-6.6.13-gentoo-x86_64-AMD_Ryzen_7_PRO_6850U_with_Radeon_Graphics-with-glibc2.39
Python version: 3.12.3
Huggingface_hub version: 0.23.4
Safetensors version: 0.4.3
Accelerate version: 0.21.0
Accelerate config: not found
PyTorch version (GPU?): 2.3.1+cu121 (False)
Tensorflow version (GPU?): not installed (NA)
Flax version (CPU?/GPU?/TPU?): not installed (NA)
Jax version: not installed
JaxLib version: not installed
Using GPU in script?:
Using distributed or parallel set-up in script?: NA

Who can help?

No response

Information

[ ] The official example scripts
[X] My own modified scripts

Tasks

[ ] An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
[X] My own task or dataset (give details below)

Reproduction

pip install accelerate==0.21.0
pip install transformers
pip install setuptools

python

[+]>>> from transformers import Trainer
[+]>>> Trainer()
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/home/dbn/src/rockfish/tmp/venv/lib/python3.12/site-packages/transformers/trainer.py", line 402, in __init__
  self.create_accelerator_and_postprocess()
File "/home/dbn/src/rockfish/tmp/venv/lib/python3.12/site-packages/transformers/trainer.py", line 4535, in create_accelerator_and_postprocess
  self.accelerator = Accelerator(**args)
                     ^^^^^^^^^^^^^^^^^^^
TypeError: Accelerator.__init__() got an unexpected keyword argument 'use_seedable_sampler'

Expected behavior

There should be no error unexpected keyword.

The requirement in setup.py for accelerate is set to >=0.21.0, but it seems that it needs to be at least 0.26.0.

This issue is mentioned in #29216 #28111

amyeroberts commented 2 weeks ago

Hi @danielnelson, thanks for raising this issue!

Indeed, we might have to update our accelerate versions cc @ydshieh

cc @muellerzr @SunMarc When does use_seedable_sampler come in?

SunMarc commented 2 weeks ago

Hi, use_seedable_sampler was introduced in this PR, so in v0.26 of accelerate. We introduced 4 months ago AcceleratorConfig which have use_seedable_sampler as arg. Then we initialize the accelerator based on this config, hence the issue. It should be fixed with the above PR @danielnelson, could you give it a try ?

huggingface / transformers