Skip to content

Latest commit

 

History

History
34 lines (29 loc) · 1.2 KB

README.md

File metadata and controls

34 lines (29 loc) · 1.2 KB

RabbitCafe

Simple AI translator backend for local and cloud AI translation

Problem of existing translation platfroms

  • Portability: most of the current product could not be deployed locally
    • or they charge you a lot
  • Secrecy: most translations, including the game you play, the document you read, should be kept secret.
  • Blurry: the translators are not yours, thus applying lora on it is difficult

RabbitCafe

** WARNING - lots of rabbits ** the product vision is to create, merge, and serve huge number of loras sitting on base models
🐇 Serious, Percise, Happy, ???
Pick your selection and start translating
Host very different fused-models and let the a filter decide who to translate what

Current Progress

  • allows using mt5-game model straight from the bin
    • the mt5-small game model is hosted at about 200mb vram, or using cpu (will add onnx support if requested)
  • the translation speed is ok, current quality is poor though, nothing leaves your local network
  • currently loras are not implemented, will be released soon

Install

Linux install after cloneing

python3 -m venv venv
source venv/bin/activate
pip install -f requirements.txt

Linux run

python app.py