Dataset Card for toanmath.com-full

Dataset Description

The toanmath.com-full dataset is a massive, high-quality collection of mathematical documents and images sourced from ToanMath.com, the leading Vietnamese repository for high school mathematics materials.

With a total size of 194 GB, this dataset is designed for large-scale training of Vision-Language Models (VLM), OCR systems, and Mathematical Language Models specialized in the Vietnamese language and curriculum (Grades 10-12).

• Repository: daominhwysi/toanmath.com-full
• Language: Vietnamese
• Total Size: ~194 GB
• Format: Parquet (sharded) and Raw Zip (PDFs/Images)

Data Types & Modalities

• Text: Mathematical problems, solutions, and explanations (LaTeX/text).
• Images: Scanned exam papers, geometric diagrams, and mathematical plots.
• Documents: Full PDF files of Vietnamese national exams and mock tests.

Dataset Structure

The dataset is organized into two main parts: processed parquet files for direct training and raw zip files for archival/custom processing.

1. Parquet Chunks

There are 315 chunks (chunk_0.parquet to chunk_314.parquet), each approximately 525 MB. They contains converted images from the raw pdf files.

2. Archive Files (.zip)

• toanmath_data_pdf.zip: The primary collection of original PDF documents.

Applications

This dataset is ideal for:

1. Math OCR: Training models to recognize Vietnamese text mixed with complex LaTeX formulas.
2. Layout Analysis: Detecting mathematical blocks, headers, and figures in educational PDFs.
3. VLM Pre-training: Teaching models like LLaVA or Qwen-VL to solve Vietnamese math problems from images.
4. Fine-tuning LLMs: Improving mathematical reasoning in Vietnamese.

Dataset Creation

Source Data

The data was collected from ToanMath.com, covering:

• National High School Graduation Exams (THPT Quốc gia).
• Semester exams from specialized schools (Trường chuyên) across Vietnam.
• Math competition materials (HSG).

Licensing

This dataset is intended for research and educational purposes. Users should respect the intellectual property of the original authors on ToanMath.com. It is generally recommended to use this under a CC BY-NC-SA (Creative Commons Attribution-NonCommercial-ShareAlike) framework.

Citation

If you use this dataset in your research, please credit the original source:

ToanMath - Tài liệu và đề thi Toán học (https://toanmath.com)
Compiled by daominhwysi.