---

name: latex-ocr description: "Guides LaTeX-OCR pix2tex image-to-LaTeX OCR, CLI, Python API, GUI, service API, dataset preparation, training, and troubleshooting workflows." disable-model-invocation: true metadata: disco-role: operating license: MIT

LaTeX-OCR / pix2tex Repo Skill

Use this skill when a task involves the LaTeX-OCR repository or the pix2tex Python package: converting equation images to LaTeX, operating the CLI or Python API, launching the GUI or API service, preparing OCR datasets, training or evaluating the model, or troubleshooting optional dependencies.

First Checks

1. Read references/repo-provenance.md when you need to decide whether this skill matches a current checkout.
2. For a fresh runtime, install the smallest needed extra:
   • pip install pix2tex for CLI/Python inference.
   • pip install "pix2tex[gui]" for the desktop GUI.
   • pip install "pix2tex[api]" for FastAPI/Streamlit service workflows.
   • pip install "pix2tex[train]" for dataset/training/evaluation workflows.
3. Run scripts/check_pix2tex_environment.py to check imports, package version, CLI availability, selected extras, and CPU/CUDA visibility without downloading checkpoints or running inference.
4. Read references/troubleshooting.md if an install/import/backend issue appears before choosing a sub-skill.

Route Map

• Read sub-skills/ocr-inference/SKILL.md for CLI and Python OCR inference, image preprocessing, checkpoints, --no-cuda, output rendering, and prediction-quality troubleshooting.
• Read sub-skills/interactive-apps-and-api/SKILL.md for latexocr GUI, screenshot tools, FastAPI /predict/ and /bytes/ routes, Streamlit frontend, and Docker API recipes.
• Read sub-skills/data-preparation/SKILL.md for extracting formulas, de-macroing TeX, tokenizer/dataset pickle creation, rendering LaTeX into PNGs, and safe data acquisition planning.
• Read sub-skills/training-and-evaluation/SKILL.md for model/resizer training, evaluation metrics, config files, checkpoint handling, and GPU-memory planning.

Operating Boundaries

• Do not assume the original repository checkout is available. Runtime guidance lives in this skill tree; the source paths in provenance are evidence, not links to open during normal use.
• Do not instantiate LatexOCR() unless model checkpoint downloads and compute are acceptable. Use helper scripts first when you only need environment or input validation.
• Treat GUI, Docker, Streamlit, TeX rendering, scraping, and full training as optional side-effectful workflows. Ask before starting servers, desktop installers, downloads, scraping, or long runs.
• CPU is valid for CLI/Python inference via --no-cuda or an arguments object with no_cuda=True, but it may be slower than CUDA. Do not claim CUDA works unless the target environment has a compatible CUDA PyTorch build and a successful device smoke check.

Repository Metadata

• Structured router placement is in references/repo-routing-metadata.json.
• Cross-cutting install and optional dependency guidance is in references/troubleshooting.md.