BigDocs-7.5M: Permissively-Licensed Multimodal Training Data for Documents and Code

BigDocs-7.5M is a dataset created by ServiceNow for training multimodal models on document and code tasks, as described in the associated arXiv paper. The dataset was last updated on June 20, 2025, and is hosted on Hugging Face. It appears to contain both text and image data, with some parts distributed using an image identifier column that requires a provided script to reconstruct.

Use Cases

• Training multimodal document understanding models based on combined text and image data.
• Developing models for code-related tasks using the described training data.
• Benchmarking model performance on tasks involving both visual and textual document elements.
• Fine-tuning large language models on a corpus of documents and code.

Strengths

• Dataset is associated with a peer-reviewed arXiv paper, suggesting academic rigor.
• Includes tooling (a Python script) to handle data reconstruction for parts distributed with image identifiers.
• Last updated on 2025-06-20, indicating recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

ServiceNow, as indicated in the author field.

Collection Method

Created as training data for the associated research paper 'BigDocs: An Open and Permissively-Licensed Dataset for Training Multimodal Models on Document and Code Tasks'.

Time Range

null

Freshness

Last updated 2025-06-20 17:48:32.

Geography

null