KoAlpaca

lastmod 2023-03-20created 2023-03-201 edits

Korean Alpaca model

Features

• Trained the same way as Stanford's Alpaca model
• Uses Polyglot-ko 5.8B and LLaMA 7B as backbone models
  • LLaMA was trained on too little Korean data, so its Korean performance is low; a Korean model was additionally trained
• LLaMA's 52k instruction dataset was translated with the DeepL API

See also

• Deep learning
• LSTM

Favorite site

• Github - KoAlpaca