python-v2.14.5

run-llama/liteparsepython-v2.14.5Sep 15, 2026by github-actions[bot]

Upgrading?

AI Summary

This release delivers the v2.14.5 updates to the Python library, bringing the same set of improvements in text extraction precision and OCR capabilities. It includes updates to the core PDFium engine and new features for handling complex document structures.

Key Highlights

  • Improved text extraction accuracy for rotated and sheared items
  • Enhanced OCR filtering for figures and charts
  • Added page label extraction and async parsing support
  • Updated to the latest PDFium library with new rendering capabilities

New Features

  • Faithful annotation/form/struct-tree consumer
  • Page label extraction
  • Async python parse support
  • Improved OCR filtering
  • Typed and borrowed accessors for page objects
  • Better text-matrix scale accuracy
  • Italic baseline geometry and widget-appearance text handling

Full Release Notes

## What's Changed
* feat(pdfium): facts a faithful annotation/form/struct-tree consumer by @logan-markewich in https://github.com/run-llama/liteparse/pull/448
* fix: report correct markdown layout bbox for rotated items by @logan-markewich in https://github.com/run-llama/liteparse/pull/450
* feat: improve OCR filtering for figures/charts by @logan-markewich in https://github.com/run-llama/liteparse/pull/453
* pdfium: set_crop_box, render_into and Bitmap::new_with_format by @logan-markewich in https://github.com/run-llama/liteparse/pull/454
* feat: page label extraction, async python parse by @logan-markewich in https://github.com/run-llama/liteparse/pull/455
* chore: update readme with benchmarks by @logan-markewich in https://github.com/run-llama/liteparse/pull/456
* pdfium: expose page objects as typed, borrowed accessors by @logan-markewich in https://github.com/run-llama/liteparse/pull/457
* raw_text: form the text-matrix scale with better accuracy by @logan-markewich in https://github.com/run-llama/liteparse/pull/458
* raw_text: italic baseline geometry, widget-appearance text, soft-hypen by @logan-markewich in https://github.com/run-llama/liteparse/pull/459
* raw_text: derive rotated and sheared grounding bounds from glyph outlines by @logan-markewich in https://github.com/run-llama/liteparse/pull/460
* feat: use latest pdfium lib by @logan-markewich in https://github.com/run-llama/liteparse/pull/461


**Full Changelog**: https://github.com/run-llama/liteparse/compare/crates-v2.14.4...python-v2.14.5

Related Repositories