TransGPT

TransGPT is a GPT-powered comimc translation tool for tranlators to improve their efficiency by automating certain steps in their workflow.

Tech Stack

Client: React, Next.js (13) , Chakra UI

Server: Node.js (v16.17.1), Python (3.10)

Database: PostgreSQL

Features

• GPT chat
• Glossary translation
• Translation
• OCR Text extraction
• Speech bubble detection
• DeepL translation
• Google Translate

Environment Variables

To run this project, you will need to add the following environment variables to your .env file.

Run cp .env.example .env

Read more on Environment variables in Next.js 13 here

OPENAI_API_KEY = xxxxx

NEXT_PUBLIC_SERVER_URL = xxxx
NEXT_PUBLIC_PORT = xxxxx
NEXT_PUBLIC_VERSION_DATE = Aug 2

DATABASE_URL= 

EMAIL_VERIFICATION_SECRET=
EMAIL_FROM=
EMAIL_USER=
EMAIL_PASSWORD=
EMAIL_HOST = 
EMAIL_PORT = 

SECRET_KEY=

Run Locally

Clone the project

  git clone https://github.com/John-Oula/trans-gpt

Go to the project directory

  cd trans-gpt

Install dependencies

  npm install 

Go to the server directory and install dependencies

  cd server && npm install

Install python packages

  pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple/

Start the Next.js development server

  cd .. && npm run dev

Open a new terminal and start the backend server

  cd server && node app.js

Deployment

To deploy this project run

 npm install pm2 -g
 cd trans-gpt

Build the app

  npm run build

Start pm2

  pm2 start ecosystem.config.js
  pm2 save

Usage/Examples

Translation, Chat and Glossary modules

GPT Chat functionality

Glossary translation with GPT & highlighting

Same translation without glossary

OCR Comic speech bubble text extraction translation with GPT-4

System message design interface

System message output example

In the screenshot below, an additional requirement was added to instruct GPT to give an analysis of its translations. This show's that GPT's output is greatly influenced by system message and prompt design.

Further Improvement

• More focus should be put in prompt engineering to realize better GPT output

• Improve the speed of batch OCR text extraction.

• Add an admin management panel to manage glossaries preferrebly with react-admin

• Develop a mechanism to count and control the number of tokens used, especially with the "Translate All" feature. Check this repository

• Support internationalization

• Support for .rar file upload and extraction. Currently supports .zip files

• Fine-tuning GPT. Fine-tuning lets you get more out of the models available through the Open AI's API by providing:

  • Higher quality results than prompt design
  • Ability to train on more examples than can fit in a prompt
  • Token savings due to shorter prompts
  • Lower latency requests

Authors

• John Oula

Support

For support, email [email protected].