alephpi/Texo

A minimalist SOTA LaTeX OCR model with only 20M parameters, running in browser. Full training pipeline available for self-reproduction. | 超轻量SOTA LaTeX公式识别模型，仅20M参数量，可在浏览器中运行。训练全流程代码开源，以便自学复现。

What it solves

Texo is a lightweight, open-source LaTeX OCR model designed to convert images of mathematical formulas into LaTeX code. It addresses the need for a fast, accessible, and precise tool for STEM learners and researchers who need to digitize mathematical notes.

How it works

Texo is a distilled version of the PPFormulaNet-S model, which is then fine-tuned on the UniMERNet-1M dataset. It uses a combination of a model encoder and decoder to recognize patterns in mathematical images and translate them into a closed vocabulary of LaTeX commands.

Who it’s for

This project is primarily for STEM students, AI learners, and researchers who need a fast way to convert handwritten or printed mathematical formulas into LaTeX format.

Highlights

  • Minimalist Design: Contains only 20 million parameters, making it significantly smaller than other SOTA models like UniMERNet-T (107M).
  • High Performance: Maintains comparable performance to larger models in terms of BLEU and Edit distance metrics on the UniMERNet-Test dataset.
  • Consumer-Grade Training: Can be trained on consumer-level GPUs, with a minimum requirement of 16GB GPU memory.
  • Browser-Ready: Supports running directly in the browser via a companion web project.
  • Fast Inference: Optimized for lightweight and fast inference speeds.

Related

  • Project
    SakuraMathcraft/LaTeXSnipper

    A desktop application that converts screenshots, images, and PDFs into editable LaTeX formulas and text using OCR.

  • Project
    Topdu/OpenOCR

    An open-source toolkit for general OCR, providing lightweight models for unified text, formula, and table recognition and document parsing.

  • Project
    aiptimizer/TurboOCR

    A high-speed GPU document parser that converts images and PDFs into structured Markdown, including tables and formulas, using C++, CUDA, and TensorRT.

  • Project
    robertknight/ocrs

    A Rust library and CLI tool for extracting text from images using neural network models to minimize the need for image preprocessing.

  • Dispatch
    PRX Part 3: Training a Text-to-Image Model in 24 Hours

    Hugging Face and Photoroom demonstrate a text-to-image model trained in 24 hours using 32 H200 GPUs on a $1500 budget, combining pixel-space training, token routing, and representation alignment.