Installation

Flowde requires Python 3.11 or later.

Full installation

For CPU extraction and access to all the documented features, install:

python -m pip install "flowde[all] @ git+https://github.com/EPPI-Centre/Flowde.git"

This includes PaddleOCR and CPU PaddlePaddle, OpenAI and Gemini clients, and the benchmark dependencies. PaddleOCR downloads model files when the extractor first runs.

The GPU version of PaddlePaddle needs a different installation; follow GPU support if you want extraction to use an NVIDIA GPU.

Minimal installation

If you are bringing your own processing functions, install the core package:

python -m pip install "flowde @ git+https://github.com/EPPI-Centre/Flowde.git"

Add only the optional features that you need from the following sections.

Install specific extras

PaddleOCR helpers

The paddle extra installs PaddleOCR. For CPU extraction, also install the PaddlePaddle runtime:

python -m pip install "flowde[paddle] @ git+https://github.com/EPPI-Centre/Flowde.git" paddlepaddle

For GPU extraction, use the runtime described under GPU support.

OpenAI helpers

This extra supports both OpenAI and Azure OpenAI:

python -m pip install "flowde[openai] @ git+https://github.com/EPPI-Centre/Flowde.git"

Gemini helpers

python -m pip install "flowde[gemini] @ git+https://github.com/EPPI-Centre/Flowde.git"

Benchmarking

python -m pip install "flowde[benchmark] @ git+https://github.com/EPPI-Centre/Flowde.git"

Extras can be combined. For example, install flowde[openai,benchmark] if you already have images and want to parse and evaluate them using OpenAI.

GPU support

PaddleOCR image extraction runs on your computer. An NVIDIA GPU is used only if your environment has a compatible GPU-enabled PaddlePaddle installation. The LLM helpers send requests to the selected API provider; a local GPU does not accelerate those requests.

For a GPU environment, install the Flowde extras you need without the all extra, then follow the official PaddlePaddle installation instructions to install the GPU runtime for your operating system, Python and CUDA versions. The all extra includes the CPU runtime.

After installing the GPU runtime, check the environment:

import paddle

print(paddle.is_compiled_with_cuda())
print(paddle.device.cuda.device_count())

The first value must be True, and the second must be at least 1, before using make_paddle_layout_extract_fn(device="gpu") .