I Trained A 67M-param LaTeX OCR Model That Runs On A Laptop CPU — And Built A New Style-aware Dataset To Train It. Weights, Data, And Training Code All Open (MIT).

Hey everyone! I’ve been working on a little side project I wanted to share: latex-ocr, a standalone formula OCR model — you feed it an image of a math formula, it spits out the LaTeX source.

The main hook: it’s only 67M parameters, so it runs comfortably on a laptop CPU. No GPU, no 300M-parameter monster to load. It’s a CoCa-style model (contrastive captioner adapted for OCR), and despite the small size it beats the 107M UniMER-tiny baseline and gets pretty close to the 325M one on plain formulas.

The part I’m actually most proud of is the dataset. Real papers don’t just use plain symbols — you see mathbb{R}, mathcal{F}, mathfrak{g} everywhere, and existing OCR datasets basically ignore font styles, so models trained on them can’t read (or hallucinate) those macros. So I rebuilt ~1.3M formulas with a MathJax → SVG → PDF → PNG pipeline and injected font-style macros with semantic heuristics (number sets → mathbb, vectors → mathbf, differentials → mathrm). On that styled test set it clearly outperforms all the baselines — fair warning though, those baselines are zero-shot on styled data, so take that comparison with a grain of salt. The plain-split numbers are the like-for-like ones.

Everything is open: model weights and dataset on Hugging Face, training recipes included if you want to reproduce or fine-tune it yourself, MIT license. There’s also a FastAPI server and a Gradio web UI, so you can drag-and-drop an image and see the LaTeX with a rendered preview.

Repo: https://github.com/PadishahIII/latex-ocr Model: https://huggingface.co/PadishahIIIXXX/latex-ocr Dataset: https://huggingface.co/datasets/PadishahIIIXXX/latex-ocr-dataset

Happy to answer questions about the training setup, the data pipeline, or anything else. Would love feedback — especially if you try it on your own gnarly formulas and it breaks, that’s genuinely useful.