Skip to content
Use case

Keeping footnotes when converting a paper to EPUB

Updated:

Footnotes are kept as linked EPUB notes when they sit below the body text and start with a marker that matches a superscript call. On the QLoRA paper (arXiv, CC BY 4.0) other engines recover more continuous text than ours, which the table shows.

Try 5 pages free

Original and result

Footnotes, tables and a long appendix in one column.

Original PDF
Page of the original PDF
Result (EPUB, as a reader shows it)
The same passage in the converted EPUB

Download source PDF Download the sample EPUB

Source and licence

Document
QLoRA paper (arXiv 2305.14314)
Licence
CC BY 4.0
Pages
26
Traits
text PDF, one column, tables, footnotes, academic paper, long

Dettmers, Pagnoni, Holtzman and Zettlemoyer, “QLoRA: Efficient Finetuning of Quantized LLMs”, arXiv:2305.14314v1, licensed CC BY 4.0. Pages shown unmodified; the EPUB is our conversion.

Settings used

  • Engine: EpubCC engine, build of 2026-11-03 (the one measured in the benchmark).
  • Pages: the whole document.
  • Options: none. The engine has no per-document settings yet; the same defaults apply to every file.

Validation and results

EPUBCheck EPUBCheck on our EPUB: 0 errors, 0 warnings.

8-gram recall · Like 4-gram recall with 8 words: drops when reading order or line joins break sentences (continuity). Methodology
Engine8-gram recall
EpubCC engine (EPUB output)0.786
EpubCC engine (plain text)0.782
pdf_oxide · extract_text0.817
pdf_oxide · to_plain_text (XY-cut)0.852
pdf_oxide · Markdown0.868
PDFium · text().all()0.908
OpenDataLoader PDF (local mode)0.917
pdftotext (Poppler)0.770

Limitations we observed

  • OpenDataLoader and PDFium have a higher 8-gram recall than our engine on this document.
  • Whether each note was linked is checked by hand; the benchmark measures text only.

Convert a similar PDF

Share