Datasets:

Modalities:
Text
Formats:
parquet
ArXiv:
Libraries:
Datasets
Dask
License:
InfiMM-WebMath-40B / README.md
xiaotianhan's picture
Update README.md
1f8df9b verified
|
raw
history blame
1.45 kB
metadata
license: odc-by
task_categories:
  - text-generation
language:
  - en
  - zh
tags:
  - pretrain
  - multi-modal
size_categories:
  - 10M<n<100M

InfiMM-Math Contents

The dataset is structured as follows:

{
  "URL": ...,  # document url.
  "text_list": ...,  # extracted text from document, None means a image.
  "image_list": ...,  # extracted image urls from document, None means text.
  "metadata": ...,  # JSON containing information from the extraction process.
}

Interleaved Text and Images

The text_list and image_list are designed to be parallel lists that together represent the sequential content of the document. The interleaving works as follows:

•	If an element in `text_list` is a string (i.e., not None), the corresponding element in image_list is None, indicating that this position in the document contains text.
•	If an element in `text_list` is None, the corresponding element in image_list contains an image URL, indicating that this position in the document contains an image.

This structure preserves the original order of text and images, allowing for accurate reconstruction of the document’s content flow.

License

InfiMM-Math is made available under an ODC-By 1.0 license; users should also abide by the CommonCrawl ToU: https://commoncrawl.org/terms-of-use/. We do not alter the license of any of the underlying data.

Citation Information