DataFlow-WebUI User Document
输入“/”快速插入内容
DataFlow-WebUI User Document
DataFlow-WebUI User Document
用户8461用户8461
用户9675用户9675
用户3566用户3566用户7565用户7565用户3721用户3721+2
2月1日修改
中文版：DataFlow-WebUI用户文档（中文）
Overview
DataFlow is a framework primarily designed for extracting textual data and building data governance pipelines for text data, serving large language model (LLM) training. It is organized as a Python library.
To improve usability, we have built DataFlow-WebUI, which adopts Vue and FastAPI as the frontend and backend technology stack. The backend uses Ray to manage the execution of operators and pipelines in DataFlow. With local deployment, users can directly experience DataFlow’s design philosophy and high-quality pipelines through a graphical web interface. Both the frontend and backend can be used as open-source projects, providing reference and inspiration for developing workflow-construction frameworks.
Please note that this project has the following characteristics:
1.
It is designed to serve DataFlow, with built-in support for DataFlow pipelines and built-in sample datasets, allowing users to try it out immediately after installation.
2.
DataFlow operators can be intuitively composed into pipelines through drag-and-drop interactions on the web interface. Users can run pipelines, monitor execution status in real time, and download the processed data at any time.
3.
Currently, only API-deployed large model backends are supported. For local models, users can first deploy services via vLLM or SGLang, and then configure API-based access.
4.
As a research-oriented open-source project, to ensure simplicity and ease of maintenance, business-oriented features such as user management are not included. The project mainly targets local deployment and hands-on experimentation.
Quick Start
1.
After starting the DataFlow-WebUI service locally, open http://localhost:<backend port>/ in your browser to access the interface. (In this demo, port 8123 is used; the default is 8000.)
2.
Click the "Serving Manager" option in the left sidebar to enter the LLM service configuration interface. Click "Add" and enter the service parameters as shown in the image below. Click "Confirm" to complete the configuration of your LLM service in the DataFlow-WebUI backend.
3.
Then, return to "DataFlow Hub" and click the Pipeline button. The system will display the pre-set operator pipeline templates based on LLM API calls. Click on a template to view the visualized pipeline on the canvas below. (Note: The Text2SQL pipeline requires configuring the DB Manager in advance; please refer to the section below for specific configuration details.)