Local AI Engineering with Ollama

Pulling Models

ollama pull downloads a model from the Ollama registry to your local disk. You run a model with ollama run, which pulls automatically if needed, but pulling explicitly is the right move when you want to grab models ahead of time, on a fast network, before you actually need them.

Let's start with a simple model (Granite 3.3).

ollama pull granite3.3:2b

The format is $MODEL:$TAG. The tag picks a specific variant: parameter size, quantization, instruction-tuning, or a combination.

In our case, granite3.3:2b means IBM's Granite 3.3 at roughly 2 billion parameters. Leave the tag off and you get :latest