Skip to content

Installation

Requirements

  • Python 3.10, 3.11, or 3.12
  • Tesseract OCR (brew install tesseract / apt install tesseract-ocr)

For most users — includes everything (server, embeddings, vector DB, OCR):

pip install "longparser[gpu]"

This pulls the standard PyTorch wheel from PyPI which includes CUDA support. It works on CPU machines too — torch just runs in CPU mode automatically.

Core SDK only (no server, no torch)

Just document parsing and chunking — minimal footprint:

pip install longparser

Pick only what you need

Extra Installs Notes
server FastAPI + MongoDB + Redis + LangChain chat REST API layer
embeddings-gpu sentence-transformers GPU torch recommended
embeddings-cpu sentence-transformers Pre-install CPU torch first
faiss-gpu faiss-gpu vector store Included in [gpu]
faiss-cpu faiss-cpu vector store Included in [cpu]
chroma ChromaDB vector store No torch needed
qdrant Qdrant vector store No torch needed
latex-ocr-gpu pix2tex equation OCR GPU torch
latex-ocr-cpu pix2tex equation OCR CPU torch
docx-equations OMML→LaTeX for DOCX/PPTX No torch needed
langchain LangChain core adapter No torch needed
llamaindex LlamaIndex reader adapter No torch needed
pptx python-pptx indent detection No torch needed
gpu All of the above (GPU) ✅ Recommended
cpu All of the above (CPU-only) See Advanced below 
# Example: server + chroma only (no torch needed at all)
pip install "longparser[server,chroma]"

# Example: server + GPU embeddings + qdrant
pip install "longparser[server,embeddings-gpu,qdrant]"

Advanced: CPU-only install (minimal size)

Use this if you want to save ~1.8 GB by avoiding the CUDA torch build. Useful for Docker images, edge devices, or CI environments.

# Step 1 — install CPU-only torch first (~230 MB vs ~2 GB)
pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu

# Step 2 — install LongParser with CPU extras
pip install "longparser[cpu]"

git clone https://github.com/ENDEVSOLS/LongParser.git
cd LongParser

# GPU (default)
uv sync --extra gpu

# CPU-only
uv sync --extra cpu

Docker

docker compose up

The server starts on http://localhost:8000.

Verify Installation

import longparser
print(longparser.__version__)  # 0.1.2