Starstruk Logo MarkStarstruk

v0.17.0

jdx/misev0.17.0Sep 23, 2025by tarun-menta

AI Summary

Introduces a new architecture for layout processing with a trained-from-scratch model and tokenizer updates.

Key Highlights

  • New unified tokenizer
  • New layout model trained from scratch
  • Significant improvements across many domains

New Features

  • New unified tokenizer
  • New layout model architecture

Full Release Notes

## Releasing new layout model
- Moving to a new architecture for layout, trained from scratch. Significant improvements across many domains

## What's Changed
* feat: new unified tokenizer by @VikParuchuri in https://github.com/datalab-to/surya/pull/442
* Layout Model Release by @tarun-menta in https://github.com/datalab-to/surya/pull/461
* Dev by @tarun-menta in https://github.com/datalab-to/surya/pull/463


**Full Changelog**: https://github.com/datalab-to/surya/compare/v0.16.7...v0.17.0