LLM
Also called: large language model, language model
Large language model: an AI trained on huge volumes of text, able to understand a request and answer it, token by token.
LLM is short for Large Language Model. It is a Transformer-type neural network, trained on gigantic amounts of text to learn the patterns of language, code and the knowledge they contain. Conversational assistants are built on this type of model, as are open model families such as Llama, Mistral or Qwen, which you can install on your own hardware.
An LLM works by predicting the most likely continuation of a text. Your question is split into tokens, pieces of words, then the model computes the next token, adds it to the text and starts again until the answer is complete. Its knowledge is stored in its parameters, billions of numbers adjusted during training. An LLM does not look things up on the internet by itself and can be confidently wrong: always check its important answers.
To run an LLM on your own machine, the main criterion is the GPU's video memory: it must hold the model, depending on its size and quantization, plus a margin for the context. The bandwidth of that memory then sets the response speed, in tokens per second. Once quantized, a model with 7 to 14 billion parameters fits on a 16 GB card; larger models need several GPUs or professional cards.
Our AI workstations are delivered ready to use with Ollama and Open WebUI: your LLM runs locally, and your data stays with you.