Skip to content

Language model

A language model (LLM) is the AI that reads and writes text. It guesses the next word based on everything it was trained on. Domino IQ runs a language model on the server.

A language model is the part of AI that reads and writes text. Its full name is Large Language Model, or LLM for short. ChatGPT, Claude and Llama are examples.

How does a language model work?

The model does not think like a human. It splits the text into small pieces called tokens. Then it guesses the most likely next piece, again and again. It builds on patterns from everything it was trained on.

That is why it can be wrong. If it follows a wrong track, it makes up an answer. This is called hallucinating. A model that is not trained on your language often answers poorly in that language.

Large or small model

A language model is a file. It can take up anything from a few GB to several hundred GB. Models for local use often take up 16 to 32 GB. With quantisation a large model can be made smaller, so it fits on an ordinary graphics card.

A small model made for one specific task can be better than a large model that has to do everything. It is also cheaper to run.

Language models in Domino IQ

Domino IQ runs one language model at a time on the Domino server. The model must be in the GGUF format. Many open models can be downloaded for free from Hugging Face. Choose a model that fits your language, your field and your graphics card.

All terms in the glossary