Hugging Face's logo

PDFluent
/
pdf-heading-level-classifier

Tabular Classification
English
pdf
document-ai
document-understanding
layout-analysis
heading-detection
document-structure
lightgbm
rag
pdf-to-markdown
document-parsing
Model card Files
xet
 Community
pdf-heading-level-classifier / PROVENANCE.md
jasperdew's picture
jasperdew
Add heading_level_979 weights, card and attribution
f9d08ab verified
preview code
|
Raw History Blame Contribute Delete
955 Bytes

Provenance: heading_level_979

  • No base model. LightGBM, trained from scratch.
  • Weights (model.json.gz) sha256 44e777b277a9230c4cfb7cb9109ef3ba016e955112526f1b7079002853d0d747, equal to weights_sha256 in the SDK 1.3.0 source (scripts/extraction/learned/models/heading_level_979/manifest.json, tag v1.3.0). The model is compiled into the SDK; it is not a models-v1 asset.
  • Documents: arXiv (CC BY 4.0, CC0 1.0), PubMed Central (CC BY), synthetic pages made by PDFluent. CC BY-SA excluded. Per-document licence, title, URL: attribution.json. PMC author names: PMC-AUTHORS.json.
  • Split and counts: manifest.json. Leak check against the holdout registry: see manifest.json and parity.json.
  • 17 PMC papers have source XML whose hash differs from the training manifest. Article identifiers match. Listed in PROVENANCE-CHECK.json (internal; not uploaded).
  • Page copyright of third-party pages used as training pages is not cleared.