README.md

metadata

annotations_creators:
  - no-annotation
language_creators:
  - crowdsourced
language:
  - en
license:
  - cc-by-sa-3.0
  - gfdl
multilinguality:
  - monolingual
size_categories:
  - 10M<n<100M
source_datasets:
  - original
task_categories:
  - fill-mask
task_ids:
  - masked-language-modeling
pretty_name: Team-PIXEL/rendered-wikipedia-english
tags:
  - masked-auto-encoding
  - rendered-language-modeling
  - rendered-language-modelling

Dataset Card for Team-PIXEL/rendered-wikipedia-english

Dataset Description

• Homepage: https://github.com/xplip/pixel
• Repository: https://github.com/xplip/pixel
• Paper: Language Modelling with Pixels
• Point of Contact: Phillip Rust
• Size of downloaded dataset files: 125.66 GB
• Size of the generated dataset: 125.56 GB
• Total amount of disk used: 251.22 GB

Dataset Summary

This dataset contains the full English Wikipedia from February 1, 2018, rendered into images of 16x8464 resolution.

The original text dataset was built from a Wikipedia dump. Each example in the original text dataset contained the content of one full Wikipedia article with cleaning to strip markdown and unwanted sections (references, etc.). Each rendered example contains a subset of one full article. This rendered English Wikipedia was used to train the PIXEL model introduced in the paper Language Modelling with Pixels by Phillip Rust, Jonas F. Lotz, Emanuele Bugliarello, Elizabeth Salesky, Miryam de Lhoneux, and Desmond Elliott.

The original Wikipedia text dataset was rendered article-by-article into 11.4M examples containing approximately 2B words in total. The dataset is stored as a collection of 338 parquet files.

It was rendered using the script openly available at https://github.com/xplip/pixel/blob/main/scripts/data/prerendering/prerender_wikipedia.py. The text renderer uses a PyGame backend and a collection of merged Google Noto Sans fonts. The PyGame backend does not support complex text layouts (e.g. ligatures and right-to-left scripts) or emoji, so occurrences of such text in the Wikipedia data have not been rendered accurately. Each example consists of a "pixel_values" field which stores a 16x8464 (height, width) grayscale image containing the rendered text, and an integer value "num_patches" which stores how many image patches (when splitting the image into 529 non-overlapping patches of resolution 16x16 pixels) in the associated images contain actual text, i.e. are neither blank (fully white) nor are the fully black end-of-sequence patch.

You can load the dataset as follows:

from datasets import load_dataset

# Download the full dataset to disk
load_dataset("Team-PIXEL/rendered-wikipedia-english", split="train")

# Stream the dataset directly from the hub
load_dataset("Team-PIXEL/rendered-wikipedia-english", split="train", streaming=True)

Dataset Structure

Data Instances

• Size of downloaded dataset files: 125.66 GB
• Size of the generated dataset: 125.56 GB
• Total amount of disk used: 251.22 GB

An example of 'train' looks as follows.

{
    "pixel_values": <PIL.PngImagePlugin.PngImageFile image mode=L size=8464x16
    "num_patches": "469"
}

Data Fields

The data fields are the same among all splits.

• pixel_values: an Image feature.
• num_patches: a Value(dtype="int64") feature.

Data Splits

train
11446535

Dataset Creation

Curation Rationale

More Information Needed

Source Data

Initial Data Collection and Normalization

More Information Needed

Who are the source language producers?

More Information Needed

Annotations

Annotation process

More Information Needed

Who are the annotators?

More Information Needed

Personal and Sensitive Information

More Information Needed

Considerations for Using the Data

Social Impact of Dataset

More Information Needed

Discussion of Biases

More Information Needed

Other Known Limitations

More Information Needed

Additional Information

Dataset Curators

More Information Needed

Licensing Information

Most of Wikipedia's text and many of its images are co-licensed under the Creative Commons Attribution-ShareAlike 3.0 Unported License (CC BY-SA) and the GNU Free Documentation License (GFDL) (unversioned, with no invariant sections, front-cover texts, or back-cover texts).

Some text has been imported only under CC BY-SA and CC BY-SA-compatible license and cannot be reused under GFDL; such text will be identified on the page footer, in the page history, or on the discussion page of the article that utilizes the text.

Citation Information

@article{rust-etal-2022-pixel,
  title={Language Modelling with Pixels},
  author={Phillip Rust and Jonas F. Lotz and Emanuele Bugliarello and Elizabeth Salesky and Miryam de Lhoneux and Desmond Elliott},
  journal={arXiv preprint},
  year={2022},
  url={https://arxiv.org/abs/2207.06991}
}

Contact Person

This dataset was added by Phillip Rust.

Github: @xplip

Twitter: @rust_phillip