This is a command-line tool to translate EPUBs using multilingual transformer models. Processing runs locally and requires a CUDA-compatible GPU.
| Features |
|---|
| 200+ languages |
| Choose from several models |
| Unrestricted offline translation |
| Create bilingual ebooks |
| Retains ebook formatting |
Currently, the output quality is comparable to Google Translate and DeepL.
You should test the model on a sample before translating an entire book.
A test sample is included in tests/resources/.
In general, the quality is better between closely related languages.
Models
These are the latest supported models. Quantized versions are used to reduce the size and RAM requirements.
| Model | Parameters | Languages | Size | Technology |
|---|---|---|---|---|
| NLLB200 | 1.3B | 200 | 1.3 GB | CTranslate2 |
| small100 | 330M | 100 | 0.6 GB | SentencePiece, bitsandbytes |
AI evolves quickly. You are welcome to suggest a new state-of-the-art multilingual model that could be added.
Prerequisites
- CUDA graphics card (Nvidia)
- python
Setup (first time)
- download the source code from a release, move it to your preferred location
- open terminal in the unzipped folder
pip install -r requirements.txt- downloads dependenciespython ui.pyorpython3 ui.py- starts the program
Update to new version
- delete the (release) folder
- download a new release, move it to your preferred location
- open terminal in the unzipped folder
- optionally:
pip install -r requirements.txt- updates dependencies
Uninstall
- check where the program data is located (About menu in the program)
- delete both the program data and source code folders
- optionally: uninstall pip dependencies in requirements.txt
Feedback
I appreciate any bug reports: EPUB formatting, translation, installation etc. Please use the issue tracker or discussions.