This section describes the two common ways to install DataFlow.
Which one you choose depends on whether you are a user or a developer.

• General users (recommended): You just want to use existing pipelines and operators
  → Install the stable release from PyPI
• Developers / early adopters: You want the latest features or plan to contribute to DataFlow
  → Clone from GitHub and install in editable (dev) mode

We recommend using uv for installation.

---

Option 1: General Users (PyPI Stable Release)

Recommended for quick start and day-to-day usage, with no need to modify the source code.

Install

API / CPU only:

pip install uv
uv pip install open-dataflow

Local GPU inference (choose your backend):

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

pip install uv
uv pip install open-dataflow[sglang]

DataFlow requires Python ≥ 3.10. GPU-related dependencies may vary with vLLM or SGLang versions.

Verify Installation

dataflow -v

Example output (version may differ):

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

---

Option 2: Developers (GitHub Clone + Editable Install)

Recommended if you want to modify the source code, develop new operators, or submit pull requests. With this setup, any local changes take effect immediately.

Install

CPU only:

git clone https://github.com/OpenDCAI/DataFlow
cd DataFlow

pip install uv
uv pip install -e .

Local GPU inference (choose your backend):

pip install uv
uv pip install -e .[vllm]

pip install uv
uv pip install -e .[sglang]

Python ≥ 3.10 is required.

Verify Installation

dataflow -v

---

Check Runtime Environment (All Install Methods)

Regardless of how you install DataFlow, you can inspect your current software and hardware environment with:

dataflow env

Example output:

- dataflow version: 1.0.5
- Python version: 3.10.10
- PyTorch version: 2.6.0 (GPU)
- GPU type: MetaX C500
- vLLM version: 0.8.5
- Git commit: 2135405b509a72cd11beed7be5f29ce50274d288

---

Which Should You Choose?

• ✅ Just use DataFlow → PyPI install (simple & stable)
• 🛠️ Develop, extend, or contribute → GitHub clone + editable install

Pick the option that fits your workflow and start building with DataFlow 🚀