Model

The AI brain itself. It is a big file of numbers that turns your words into an answer.

Weights

The numbers inside a model. When people say open weights, they mean you can download these numbers.

Parameters

Another word for weights. A 7B model has about 7 billion of them. More usually means smarter but bigger.

Open source

A license approved by the Open Source Initiative, like Apache 2.0 or MIT. You can use, change, and share it.

Open weights

You can download the model, but the maker sets its own rules. It is not the same as open source.

Fully open

The maker shares the weights, the code, and the training data.

License

The legal rules for what you may do with a model.

Gated

The download page asks you to sign up and accept terms first.

Hugging Face

A big website where makers post AI models for download.

GGUF

A file type that LM Studio and similar apps can run on a normal computer.

Quantization

Shrinking a model by storing its numbers with less detail. A 4-bit file is about a quarter of the full size.

Q4_K_M

A common 4-bit file choice. It is a good mix of size and quality.

RAM

Your computer's working memory. The model has to fit in it to run.

VRAM

Memory on a graphics card. Models run fastest when they fit here.

Unified memory

On Apple chip Macs, the main chip and graphics share one pool of memory.

Context window

How much text a model can keep in mind at once, counted in tokens.

Token

A small chunk of text, about three quarters of a word.

Mixture of experts

A model built from many smaller parts. Only some parts work on each word, so it runs faster.

Reasoning model

A model that thinks step by step before it answers. It is slower but better at math and logic.

LM Studio

A free desktop app for finding and running local AI models.

Hermes Agent

A tool from Nous Research that lets an AI model take actions, like using files and tools. It can use LM Studio as its engine.

Local AI

AI that runs on your own computer, so your words do not leave your machine.