pyproject.toml

[project]
name = "pdf-granite"
version = "0.1.0"
description = "CUDA-accelerated PDF -> Markdown/HTML converter using Docling + Granite Vision chart extraction"
readme = "README.md"
license = "MIT"
license-files = ["LICENSE"]
requires-python = ">=3.13"
dependencies = [
    # Docling brings docling-core, a CPU torch/torchvision, transformers, and rapidocr (OCR).
    # The Granite Vision chart-extraction path works with plain docling on >=2.107.
    # For GPU, install a CUDA torch over this venv (see README) and run with `uv run --no-sync`.
    "docling>=2.107.0",
    # OCR engine backend for RapidOCR. Without it, RapidOCR falls back to an unsupported
    # torch backend ("Unsupported configuration: torch.PP-OCRv6.det.small"). CPU-only; the
    # torch vision/layout models are the GPU win, OCR stays on CPU.
    "onnxruntime>=1.27.0",
]

[project.scripts]
pdf-granite = "pdf_granite.convert:main"
pdf-textract = "pdf_granite.textract:main"

[dependency-groups]
dev = [
    "pytest>=8.3.0",
]
textract = [
    # High-level Amazon Textract wrapper; pulls in boto3. Optional: only needed
    # to actually run pdf-textract, not to import the module or run unit tests.
    "amazon-textract-textractor>=1.8.0",
    # Loads AWS creds from the project .env into the environment for boto3.
    "python-dotenv>=1.0.0",
]

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[tool.hatch.build.targets.wheel]
packages = ["src/pdf_granite"]