This section introduces how to install DataFlow-KG. If you simply want to quickly use the pipelines and operators provided in DataFlow-KG, please refer to General User Installation to install the stable release version. If you want to contribute code to the DataFlow repository as a developer and submit Pull Requests, please refer to Developer Installation to install the dev version from the repository.

Note: For Knowledge Graph (KG) tasks, the current dependencies have been significantly streamlined and optimized. Core graph structure processing and visualization rely on networkx and pyvis, making the overall installation process more lightweight and faster. To speed up the installation, we highly recommend using uv.

1. General User Installation

If you don't have a GPU for a local inference environment and only need to use APIs and CPU functions, configure your Python environment and install the DataFlow-KG official release👇

conda create -n dfkg python=3.10 
conda activate dfkg

pip install uv
uv pip install dataflow-kg

If you want to use a local GPU for inference, you need to use the following command:

conda create -n dfkg python=3.10 
conda activate dfkg

pip install uv
uv pip install dataflow-kg[vllm]

DataFlow-KG supports Python >= 3.10 environments.

You can use the following command to verify if the installation is correct:

dfkg -v

If the installation is successful and DataFlow is the latest Release version, you will see:

open-dataflow-kg codebase version: 0.0.2
        Checking for updates...
        Local version:  0.0.2
        PyPI newest version:  0.0.2
You are using the latest version: 0.0.2.

Additionally, there is the dfkg env command to view the current hardware and software environment, which is useful for bug reporting.

2. Developer Installation

DataFlow-KG developers can install via the following commands:

If local GPU inference is not required:

conda create -n dfkg python=3.10
conda activate dfkg

git clone https://github.com/OpenDCAI/DataFlow-KG.git
cd DataFlow-KG
pip install -e .

If local GPU inference is required:

conda create -n dfkg python=3.10
conda activate dfkg

git clone https://github.com/OpenDCAI/DataFlow-KG.git
cd DataFlow-KG
pip install -e .[vllm]

You can use the following command to verify if the installation is correct:

dfkg -v

If the installation is successful and DataFlow-KG is the latest Release version, you will see:

open-dataflow-kg codebase version: 0.0.2
        Checking for updates...
        Local version:  0.0.2
        PyPI newest version:  0.0.2
You are using the latest version: 0.0.2.

Additionally, there is the dfkg env command to view the current hardware and software environment, which is useful for bug reporting.

This way, any modifications you make locally to the DataFlow-KG package will be updated in your Python environment in real-time, making development easier. Once development is complete, you can also submit a PR to contribute your new operators and pipelines to the main repository.