Token classification using LayoutXLM

[manangandhi7]

Model : LayoutXLM

I am able to use LayoutLMv2ForTokenClassification for my problem. Since the data I am using is in German, I would like to use LayoutXLM model. Unfortunately I do not find LayoutLXLMForTokenClassification on Transformers. How could I do Token classification using LayoutXLM?

To Reproduce
Steps to reproduce the behavior:

from transformers import LayoutXLMForTokenClassification
Expected behavior
ImportError: cannot import name 'LayoutXLMForTokenClassification' from 'transformers' (/opt/conda/lib/python3.7/site-packages/transformers/init.py)

Platform: Linux
Python version: 3.7

[NielsRogge]

Hi,

LayoutXLM is identical to LayoutLMv2. There's no LayoutXLM models defined in the library. You can just use:

from transformers import LayoutLMv2ForTokenClassification

model = LayoutLMv2ForTokenClassification.from_pretrained("microsoft/layoutxlm-base")

[manangandhi7]

Hi NielsRogge, many thanks for quick answer.
I get following error when I do that. Reproducing this is very simple.

from transformers import LayoutLMv2ForTokenClassification
model = LayoutLMv2ForTokenClassification.from_pretrained('microsoft/layoutxlm-base', num_labels=10)

You will get following error:

---

TypeError: init() got an unexpected keyword argument '_configuration_file'

[NielsRogge]

Which version of Transformers are you on?

[manangandhi7]

python 3.7
transformers 4.16.2
torch 1.10.2+cu102

Do you recommend any specific version of transformers?

[manangandhi7]

Issue is resolved when I downgrade transformers to v4.15.0

[NielsRogge]

Thanks, I could reproduce the issue and have reported it. We will fix it.

[NielsRogge]

Update: fixed per huggingface/transformers#15629.

We will do a patch release on Monday. For now, you can install Transformers from master: pip install git+https://github.com/huggingface/transformers.git.

Will close this issue.