Starstruk Logo MarkStarstruk

v1.8.3

datalab-to/markerv1.8.3Aug 4, 2025by VikParuchuri

AI Summary

Replaced the OCR model with a more accurate one and updated processing heuristics.

Key Highlights

  • Switched to a new OCR model better suited for math
  • Prioritized accuracy over speed
  • Deprecated the `format_lines` function

Breaking Changes

  • Dropped `format_lines` functionality

New Features

  • New OCR model
  • Improved OCR heuristics

Full Release Notes

- New OCR model that is better all-around, but particularly at math
- Improved OCR heuristics, will now prioritize accuracy over speed
- Drop `format_lines`, since `force_ocr` is generally a bit more accurate, and less error-prone

## What's Changed
* feat: allow option to keep tables split across pages by @zanussbaum in https://github.com/datalab-to/marker/pull/813
* New OCR Model by @tarun-menta in https://github.com/datalab-to/marker/pull/820
* Bump surya version by @VikParuchuri in https://github.com/datalab-to/marker/pull/821


**Full Changelog**: https://github.com/datalab-to/marker/compare/v1.8.2...v1.8.3