Popular Ollama Models: Picking the Right One

With hundreds of models in the registry, it can be overwhelming to choose. Let's look at the "Mount Rushmore" of open-weights models and where they excel.

1. Meta’s Llama 3 (The Heavyweight Champion)

Llama 3 is the most famous open-weights model in the world.

Sizes: 8B and 70B (and 405B in the cloud).
Strengths: Best general reasoning, massive training dataset, excellent at following complex instructions.
Best for: Everything. If you don't know which model to use, start with llama3.

2. Mistral / Mixtral (The Efficiency Kings)

Hosted by Mistral AI in France, these models punch far above their weight.

Sizes: 7B (Mistral) and 8x7B (Mixtral).
Strengths: "MoE" (Mixture of Experts) architecture allows the model to be fast like a small model but smart like a large one.
Best for: Rapid-fire chat, coding, and creative writing.

3. Google’s Gemma 2 (The Academic)

Built using the same technology as Google's Gemini.

Sizes: 2B, 9B, and 27B.
Strengths: Excels at math, logical reasoning, and structured data (like JSON).
Best for: Scientific data processing and logic-heavy workflows.

4. Microsoft’s Phi-3 (The Tiny Prodigy)

Microsoft proved that size isn't everything.

Sizes: Mini (3.8B), Small (7B), Medium (14B).
Strengths: Extremely small. It can run on a high-end smartphone or a very old laptop.
Best for: Devices with very low RAM (like a Raspberry Pi or a browser).

5. Specialist Models

Sometimes you need a specific tool for a specific job:

Codellama / Deepseek-Coder: Trained specifically on millions of lines of code.
Llava: A "Multimodal" model. You can send it an image and it can describe it.
Command R: Optimized for RAG (Retrieval Augmented Generation) and long-context windows.

Model Recommendation Table

Task	Recommended Model	RAM Needed
Daily Chat	`llama3`	8GB+
Writing Code	`deepseek-coder`	8GB+
Logic/Math	`gemma2:9b`	16GB+
Image Analysis	`llava`	8GB+
Old Laptop	`phi3:mini`	4GB+

Practice Exercise

Try running a "specialist" model. If you are a developer, pull deepseek-coder and ask it to write a Python script for a simple web scraper. Note how much faster (or smarter) it is at coding than the general Llama 3 model.

Key Takeaways

Llama 3 is the best all-rounder.
Mistral is known for speed and creative writing.
Gemma 2 is Google's logic-focused offering.
Phi-3 is for resource-constrained environments.
Use Specialist models for coding and vision to get better results than general models.

Module 3 Lesson 3: Popular Ollama Models