EvalSafe Invoice Processing Dataset

It is published on Hugging Face as typesafeAI/evalsafe-invoice-processing. The card lists the Datasets library plus pandas and Polars as supported libraries, and the files are distributed in parquet format, so it can be pulled with the standard Hugging Face datasets tooling or read directly from parquet.

The dataset card lists apache-2.0 as the license. Check the card on Hugging Face for the current terms before redistributing or using the data in a product.

Records combine raw invoice text with structured tabular fields. Visible examples include line items with description, quantity, unit, unit price, and amount, alongside the full raw_text of the invoice document.

The Hugging Face card reports a size in the 1K to 10K range. That places it in evaluation-scale territory rather than as a large pretraining corpus.

The card lists English as the language and tabular plus text as the modalities. The invoice examples shown are English-language documents with pound sterling amounts.

It is published under the TypeSafe AI organization on Hugging Face, the same organization behind Jev. The dataset page shows the organization account and a small number of likes and followers.

It is scoped to invoice-processing evaluation workflows. If you need other document types, non-English invoices, or a general-purpose training corpus, the card does not indicate that coverage, so look elsewhere for those uses.