The Beginners Guide to Ollama

At first glance, Ollama might seem like just another way to run language models. But spend five minutes with it, and you'll realize it’s something more: a clean, no-fuss experience for running large language models locally — without needing to be a machine learning engineer.

Whether you're building a CLI tool, prototyping a chat assistant, or just want ChatGPT-like power without the cloud, Ollama is the easiest way to get started.

Installing Ollma

Installing Ollama on macOS or Windows is dead simple. Head over to ollama.com and grab the installer for your operating system. The install process takes just a few minutes — no Python environments, no dependency errors, no drama.

Once installed, Ollama runs quietly in the background as a local service. You can interact with it via the Terminal, Command Prompt, or through its HTTP API.

Downloading Your First Model

Ollama supports a wide range of models out of the box, including LLaMA 3, Mistral, Gemma, and more. Today, let’s install Deepseek R1, one of the newer, code-savvy models.

Launch your Terminal (or Command Prompt on Windows), then run:

ollama run deepseek-r1

This command does two things:

1. Downloads the model if you don’t already have it locally.
2. Launches an interactive chat session with the model once it's ready.

No API keys, no cloud latency — you're talking to a powerful LLM, right from your machine.

What Just Happened?

Behind the scenes, Ollama handled:

• Pulling the model weights and setting up the runtime.
• Quantizing the model (to keep memory usage reasonable).
• Launching the chat interface in your terminal.

Once it’s downloaded, you can run the model again instantly — no waiting. Ollama keeps models stored locally, so switching between them is just a command away.

Want to try another model? Just run:

ollama run llama3.3

Or list all available models with:

ollama list

Looking for something more creative or opinionated? You can also customize models using Modelfiles — but we’ll save that for a later guide.

Useful Ollama Commands to Know

Once a few models are installed, these commands make it easy to manage, switch between, and customize them:

ollama list           # Show installed models
ollama pull llama3.3  # Download a specific model
ollama run llama3.3   # Start a session with a model
ollama create         # Create a custom model
ollama serve          # Start the API server manually
ollama rm llama3.3    # Remove a model to free up space

Understanding Parameter Counts (7B vs. 13B vs. 40B)

When browsing models on Ollama, it’s common to see names like llama2:7b or mixtral:8x7b. These numbers refer to parameter counts — the size of the model in billions of parameters.

Parameters are like the “knobs” the model adjusts during training to learn patterns in language. The more parameters, the more complex the model’s understanding can be.

Here’s a quick way to think about it:

• 7B (7 billion parameters): A good balance between speed and performance. Runs well on most modern laptops. Great for everyday tasks like answering questions, writing content, or basic coding help.
• 13B and up: Larger models with more detailed reasoning and richer responses. They can be slower and may need more memory to run smoothly.
• 40B+ models: Very large models that can offer more advanced outputs, but usually require a high-end machine and aren’t as fast.

🧠 Tip: For most local workflows, a 7B model is a great balance between performance and speed. Larger models may be slower or require a machine with more memory.

Where does Ollama Store Models?

Ollama stores downloaded models in a hidden folder inside your home directory: ~/.ollama/

This is where all model files live, so if you're looking to manage disk space or inspect what's been downloaded, that’s the place to check.

Deleting an Ollama Model

To remove a model and free up space on your hard drive, run:

ollama rm your_model_name

Not sure of the exact model name? Use:

ollama list

Next Steps with Ollama

• Browse the catalog of Models.
• Read the docs.
• Write your first Modelfile.