Meta's open-source language model powering thousands of apps. When Korean users type 라마 (ra-ma) on a QWERTY keyboard still set to English, the keys produce fkak. This page explains the mapping and how to make the most of Llama as a Korean user.
On the standard Korean 두벌식 (2-set) layout, typing while the input is still in English mode maps each Korean character to a specific QWERTY key. The Korean name 라마 is typed as these keys:
Put together: 라마 typed in English mode spells fkak. Korean users searching for Llama who forget to press the 한/영 key first land on this search term — and Google's intent matching now routes many of these searches to Llama results.
Llama is Meta's open-source large language model family. Llama 3 models range from 8 billion to 405 billion parameters and are freely available for commercial and research use. Because the weights are open, developers can fine-tune Llama on custom datasets, deploy it locally, and build products without API costs or usage restrictions imposed by closed models.
Korean AI startups and research labs at KAIST, SNU, and POSTECH use Llama as a base model for Korean-specific fine-tuning. Several Korean companies have released Llama-based models trained on Korean corpora, including legal, medical, and financial text. For startups, Llama eliminates the OpenAI API cost which can become significant at scale, and keeps sensitive Korean data on-premises.