Skip to content

Latest commit

 

History

History
77 lines (60 loc) · 4.77 KB

README_EN.md

File metadata and controls

77 lines (60 loc) · 4.77 KB
Raw

中文|ENGLISH

Introduction

This is an open-source tool that allows businesses to easily import their own data using Excel. The tool can be run in a company's private cloud environment to prevent sensitive data from leaking. It offers two mainstream AI models for businesses to choose from, and future updates will include security verification features to make AI chat services more secure and reliable. For more information, please refer to the introduction website of this tool.

This tool is supported by the Administration for Digital Industries (ADI), Ministry of Digital Affairs, Republic of China (Taiwan).

Table of Contents

  • System Requirements
  • Installation
  • Training
  • Inference

System Requirements

Regarding hardware, you can opt for either local resources or cloud-based computing resources, as detailed below.

On-Premises Hardware

Minimum Requirements Recommended Requirements
CPU 16 core 32 core
Memory 128 GB 128 GB
GPU Nvidia A100 40GB * 1 Nvidia A100 80GB * 2
Storage 256 GB 512 GB
Price Approximately 0.7 million NTD Approximately 1.3 million NTD

Cloud-based

Specifications (Recommended) Number of Instances Price (Per Day)
AWS p4d.24xlarge 1 Approximately 15,000 NTD
GCP a2-ultragpu-2g 1 Approximately 8,000 NTD
Azure NC48ads A100 v4 1 Approximately 8,000 NTD
  • For the latest prices, please refer to the official websites

Training Data Format

Using the question-answering content from the Pregnancy Health Education Handbook provided by the Health Promotion Administration, Ministry of Health and Welfare, Republic of China (Taiwan) as an example, the format is illustrated in data/raw_data.xlsx. Each row represents a question-answering pair, and the column definitions are as follows:

Column Name Column Definition Column Example
Q Question At what stage of pregnancy should I have a high-level ultrasound scan?
A Answer It is recommended to have it between the 20th and 24th week of pregnancy.
REF Reference High-level ultrasound scans provide clear images of the fetus and are recommended between the 20th and 24th weeks of pregnancy. This is a self-funded examination, and pregnant women can decide whether to proceed based on their own circumstances. High-level ultrasound scans cannot be used to determine the gender of the fetus. If the pregnant woman feels unwell or the fetus has special conditions, she can discuss the frequency and number of examinations with a doctor. When fetal abnormalities are found, please consult relevant specialists.

Quick Start

On an Ubuntu 20.04 TLS operating system, execute the following command in the terminal:

bash run_service.sh -i 127.0.0.1 -d ./data/raw_data.xlsx -t {your HuggingFace access token}

Where:

  • -i: The IP address of the machine, defaulting to localhost. If the web interface needs to be accessed by external machines, an externally accessible IP address should be provided.
  • -d: Path to the training data storage
  • -t: HuggingFace access token (Please visit the HuggingFace documentation page to familiarize yourself with the relevant guidelines and apply for permission.)
  • -s: skip train/inference stage (-s skip-train/skip-inference)

This command will automatically install the environment and train the model. After training is complete, you'll see the following message on the screen:

Please enter the URL link below into your browser to activate the dialog interface
    URL link: http://127.0.0.1/gpt/qa/

Copy and paste the URL link into your browser's address bar to begin the question-answering process.

Starting Only the Web Service

Execute the following command in the terminal:

bash run_service.sh -i 127.0.0.1 -d ./data/raw_data.xlsx -t {your HuggingFace access token} -s skip-train

This command will skip the training phase and directly start the web service using the existing model.

Stopping the System Service

Execute the following command in the terminal:

bash stop_service.sh