jasperdew's picture
Add formula-own-v1 weights, card and provenance
c6a2deb verified
Raw History Blame Contribute Delete
2.2 kB
PDFluent model package models-v1 -- formula model notice
=========================================================
This text accompanies the optional "formula" model package, installed with
`pdfluent models install formula`. It is a separate asset of the models-v1 release; the
release's MODELS-NOTICE.txt is unchanged (released installers pin its digest).
formula-own-v1 -- formula recognition model (formula regions as LaTeX)
----------------------------------------------------------------------
Files: formula-own-v1-encoder.onnx 18,410,860 bytes
SHA-256 37d79b54dbfffd130da66280817a245c7641aa919b1382cbab8c716d10837042
formula-own-v1-decoder.onnx 18,198,610 bytes
SHA-256 2f469643444d9a4971102ea2e860eebf837985eb4f3c769b9d3cae6b51d4ac2b
formula-own-v1-vocab.json 12,194 bytes
SHA-256 a578606406abca3528ff85c6352f92ef06a21138f7012f63956edead2afa0d03
formula-own-v1-PROVENANCE.md 770,711 bytes
SHA-256 3bda55dc817c0843db1c2ff50a1b208c5be52bab829763beacbdde92f91f118b
Copyright (c) 2026 Innovation Trigger B.V.
Trained by PDFluent from scratch; no pretrained weights. Licensed under the PDFluent SDK
licence terms: the GNU Affero General Public License v3.0 or, at your option, the PDFluent
Commercial Licence (LICENSE and LICENSE-COMMERCIAL in this package, the same files that
accompany the PDFluent SDK).
Training data: display formulas (page crops paired with their LaTeX source) from arXiv papers
whose authors released them under CC BY 4.0 or CC0 1.0. Papers under share-alike licences were
excluded from training. formula-own-v1-PROVENANCE.md lists every training paper with its licence,
title, authors and source URL (the attribution CC BY asks for), the held-out evaluation, and the
package digests. Re-rendered training images used fonts under the GUST Font Licence (Latin
Modern, TeX Gyre) and the SIL Open Font License (STIX); no font is part of the model files.
CC BY 4.0: https://creativecommons.org/licenses/by/4.0/
CC0 1.0: https://creativecommons.org/publicdomain/zero/1.0/
Commercial redistribution: permitted under the SDK licence terms stated above.