Datasets:

puhuilab
/

ph_formula_corpus_v1

✅ July 2025 – Released the formula corpus
🔜 Aug 2025 – Upcoming release of the synthetic datasets
⏳ Sep 2025 – Scheduled release of a model achieving commercial-grade qualit PHOCR

🔧 Normalization Process

Parse formulas using the KaTeX syntax tree with customized macro definitions.
Traverse the tree with a tailored algorithm to normalize the expression of each node.

Additional normalization steps include:

Standardizing \operatorname and \operatorname* usage
Unifying variations of \textxx and \mathxx commands
Normalizing spacing across all elements

📘 Dataset Usage

def test_parquet():
    import random
    import pyarrow.parquet as pq
    s
    input_file = "/path/to/formula_corpus_v1.parquet"
    table = pq.read_table(input_file)
    print(f"Total rows: {len(table)}")

    num_rows = min(100, len(table))
    indices = random.sample(range(len(table)), num_rows)
    formulas = table.column("formula")
    for idx in indices:
        print(formulas[idx].as_py())

📚 Citation

Thank you for using PH_FORMULA_CORPUS_V1.
This dataset is developed and maintained by Puhuilab.

If you use this dataset in your research or applications, please remember to give appropriate credit.

To cite PH_FORMULA_CORPUS_V1 in academic work, please use the following BibTeX entry:

@misc{phformulacorpusv12025,
  title={PHFormulaCorpusV1: A Large-Scale Normalized LaTeX Formula Dataset},
  author={Puhui Lab},
  year={2025},
  url={https://huggingface.co/datasets/puhuilab/ph_formula_corpus_v1}
}

💡 Acknowledgment

The dataset follows the licenses of its original sources: