How to Run LLMs Locally: Complete Ollama & LM Studio Guide

Why Run LLMs Locally?

Privacy: Your data never leaves your machine
Cost: No per-token charges
Speed: No network latency
Offline: Works without internet

Hardware Requirements

Model Size	RAM Required	GPU VRAM	Examples
7B parameters	8GB	6GB	Llama 3 7B, Mistral 7B
13B parameters	16GB	10GB	Llama 3 13B
70B parameters	64GB	48GB	Llama 3 70B

Option 1: Ollama (Recommended for Developers)

Installation

# macOS/Linux
curl -fsSL https://ollama.com/install.sh | sh

# Windows - Download from ollama.com

Running Models

ollama run llama3
ollama run mistral
ollama run codellama

Best Models for Ollama

llama3.2: Meta's latest, excellent all-around
mistral: Great for general tasks
deepseek-coder: Strong coding model

Option 2: LM Studio (Best for Beginners)

Download from lmstudio.ai
Browse and download models
Load a model and start chatting

Recommended Models 2025

General Tasks:

Llama 3.2 8B: Best overall open model
Qwen 2.5: Strong multilingual support

For Coding:

DeepSeek Coder V2: Best open-source coding model
Qwen2.5-Coder: Great for Asian languages

For Limited Hardware:

Phi-3 Mini: 3.8B params, runs on 8GB RAM
TinyLlama: 1.1B params, runs on 4GB

Conclusion

Running LLMs locally has never been easier. Start with a 7B model and upgrade as your needs grow. Open-source models now provide 80-90% of GPT-4's capabilities at zero cost.