pixel-base-german

PIXEL-base-german

pixel-base-german is a PIXEL model trained on the German DBMDZ BERT Corpus.

We trained the model using the architecture and codebase proposed in the 2023 Rust et al. paper Language Modelling with Pixels.

This German model was introduced and evaluated in the paper Evaluating Pixel Language Models on Non-Standardized Languages, presented at COLING 2025.

Model description

Description from https://huggingface.co/Team-PIXEL/pixel-base

PIXEL consists of three major components: a text renderer, which draws text as an image; an encoder, which encodes the unmasked regions of the rendered image; and a decoder, which reconstructs the masked regions at the pixel level. It is built on ViT-MAE.

During pretraining, the renderer produces images containing the training sentences. Patches of these images are linearly projected to obtain patch embeddings (as opposed to having an embedding matrix like e.g. in BERT), and 25% of the patches are masked out. The encoder, which is a Vision Transformer (ViT), then only processes the unmasked patches. The lightweight decoder with hidden size 512 and 8 transformer layers inserts learnable mask tokens into the encoder's output sequence and learns to reconstruct the raw pixel values at the masked positions.

After pretraining, the decoder can be discarded leaving an 86M parameter encoder, upon which task-specific classification heads can be stacked. Alternatively, the decoder can be retained and PIXEL can be used as a pixel-level generative language model (see Figures 3 and 6 in the paper for examples).

For more details on how PIXEL works, please check the paper and the codebase linked above.

Training and evaluation data

This model has been trained renderized German text from https://huggingface.co/datasets/stefan-it/german-dbmdz-bert-corpus.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.00015
• train_batch_size: 256
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.05
• training_steps: 1500000
• mixed_precision_training: Apex, opt level O1

How to use

Before being able to use this model, it is necessary to clone and configure the codebase.

Setup

Instructions from https://github.com/xplip/pixel

This codebase is built on Transformers for PyTorch. We also took inspiration from the original ViT-MAE codebase. The default font GoNotoCurrent.ttf that we used for all experiments is a merged Noto font built with go-noto-universal.

You can set up this codebase as follows to get started with using PIXEL models:

<details> <summary><i>Show Instructions</i></summary> &nbsp;

1. Clone repo and initialize submodules

git clone https://github.com/xplip/pixel.git
cd pixel
git submodule update --init --recursive

1. Create a fresh conda environment

conda create -n pixel-env python=3.9
conda activate pixel-env

1. Install Python packages

conda install pytorch torchvision cudatoolkit=11.3 -c pytorch
conda install -c conda-forge pycairo pygobject manimpango
pip install --upgrade pip
pip install -r requirements.txt
pip install ./datasets
pip install -e .

1. (Optional) Install Nvidia Apex

git clone https://github.com/NVIDIA/apex
cd apex
pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./

1. Verify Installation on Vietnamese POS tagging

# Create a folder in which we keep the data
mkdir -p data
# Get and extract the UD data for parsing and POS tagging
wget -qO- https://lindat.mff.cuni.cz/repository/xmlui/bitstream/handle/11234/1-4758/ud-treebanks-v2.10.tgz | tar xvz -C data

python scripts/training/run_pos.py \
  --model_name_or_path="Team-PIXEL/pixel-base-finetuned-pos-ud-vietnamese-vtb" \
  --data_dir="data/ud-treebanks-v2.10/UD_Vietnamese-VTB" \
  --remove_unused_columns=False \
  --output_dir="sanity_check" \
  --do_eval \
  --max_seq_length=256 \
  --overwrite_cache

If everything is configured correctly, you should expect to see results similar to the following:

***** eval metrics *****
  eval_accuracy           =     0.8632
  eval_loss               =     1.2375

</details>

Loading our model

Once everything is set up properly, pixel-base-german can be used with the codebase. To load the model:

from pixel import PIXELConfig, PIXELForPreTraining

config = PIXELConfig.from_pretrained("amunozo/pixel-base-german")
model = PIXELForPreTraining.from_pretrained("amunozo/pixel-base-german", config=config)

Framework versions

• Transformers 4.17.0
• Pytorch 2.0.1+cu117
• Datasets 2.14.5
• Tokenizers 0.13.3

Citation

@inproceedings{munoz-ortiz-etal-2025-evaluating,
    title = "Evaluating Pixel Language Models on Non-Standardized Languages",
    author = "Mu{\~n}oz-Ortiz, Alberto and Blaschke, Verena and Plank, Barbara",
    editor = "Rambow, Owen and Wanner, Leo and Apidianaki, Marianna and Al-Khalifa, Hend and Eugenio, Barbara Di and Schockaert, Steven",
    booktitle = "Proceedings of the 31st International Conference on Computational Linguistics",
    month = jan,
    year = "2025",
    address = "Abu Dhabi, UAE",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.coling-main.427/",
    pages = "6412--6419",
}

Acknowledgements

This work was funded by the European Research Council (ERC) Consolidator Grant DIALECT 101043235; SCANNER-UDC (PID2020-113230RB-C21) funded by MICIU/AEI/10.13039/501100011033; Xunta de Galicia (ED431C 2024/02); GAP (PID2022-139308OA-I00) funded by MICIU/AEI/10.13039/501100011033/ and by ERDF, EU; Grant PRE2021-097001 funded by MICIU/AEI/10.13039/501100011033 and by ESF+ (predoctoral training grant associated to project PID2020-113230RB-C21); LATCHING (PID2023-147129OB-C21) funded by MICIU/AEI/10.13039/501100011033 and ERDF; and Centro de Investigación de Galicia ‘‘CITIC’’, funded by the Xunta de Galicia through the collaboration agreement between the Consellería de Cultura, Educación, Formación Profesional e Universidades and the Galician universities for the reinforcement of the research centres of the Galician University System (CIGUS).