license: cc-by-sa-4.0
task_categories:
- text-classification
- text-generation
tags:
- agent
- math
- Mathematics
- Large
- LLM
- large
- Math
- mathematics
- LATEX
- latex
- LaTeX
- good_quality
pretty_name: MathOCR
size_categories:
- 100M<n<1B
Dataset Card for MathOCR
Dataset Details
Dataset Description
- Curated by: Roman
- Funded by Roman: [No funding needed]
- Shared by Roman: Also made by Roman
- Language(s) (NLP): English
- License: Common Creatives Attribution Share Alike 4.0
Dataset Sources [optional]
- Repository: [https://huggingface.co/datasets/DataMuncher-Labs/LaTeXOCR]
- I don't write papers or demos
Uses
Direct Use
[Pre/Post training corpra for visual understanding of mathematics.]
Out-of-Scope Use
[Its not meant, but technically works for diffusion models.] [The only way this dataset can be misused is in] A. [Incompatible models, like TTS] B. [Copying the repo without crediting mep
Dataset Structure
[The dataset uses the format of] problem_latex|solution_latex|problem_english
Dataset Creation
Curation Rationale
[I thought there was simply a lack of mathematical OCR datasets]
Source Data
Data Collection and Processing
[Data was sourced from python...] Its synthetic data
Who are the source data producers?
[Roman] [Ai assisted in making some parts of the script (sympy is hard)]
Personal and Sensitive Information
[No PII or Sensitive information] Also why I make synthtic data
Bias, Risks, and Limitations
[Doesnt cover all equation types] [Doesnt cover all numbers possible] [There are aready similar datasets]
Recommendations
Just don't use it for the wrong models. I know someone was thinking of giving this to a speech model...
Citation [optional]
BibTeX:
@dataset{datamuncher_latexocr_2026, author = {DataMuncher-Labs}, title = {LaTeXOCR}, year = {2026}, publisher = {Hugging Face}, url = {https://huggingface.co/datasets/DataMuncher-Labs/LaTeXOCR}, note = {Synthetic LaTeX equation OCR dataset} }
APA:
DataMuncher-Labs. (2026). LaTeXOCR [Dataset]. Hugging Face. https://huggingface.co/datasets/DataMuncher-Labs/LaTeXOCR
Dataset Card Authors
[Roman]
Dataset Card Contact
[Email is romanfinal@proton.me please don't spam] [Feel free to open a discussion on the repo]