Understanding Ollama: Bringing Large Language Models Home to Your Computer

"Get up and running with large language models locally."

Want to run large models on your own computer? Sounds cool, but worried about complicated configurations and high hardware requirements? Don't worry, Ollama is here to solve these problems!

Since its inception in June 2023, Ollama has rapidly gained popularity and is now the preferred tool for many to play with large models locally. With increasing attention to data privacy and edge computing, Ollama is poised to shine on more personal devices and small servers, truly bringing the power of AI large models "to the masses."

So, what exactly is Ollama?

Simply put, as its slogan suggests, Ollama is an open-source tool that allows you to easily launch and run various large language models on your own computer.

You don't need to be a configuration master; just follow the simple installation guide, often with a single command, you can run an open-source large model (like Llama 3.1, Gemma 2, etc.). It provides a clean command-line interface and background service, especially suitable for developers or enthusiasts who want to build applications based on large models. Downloading models, running them, managing them? Ollama takes care of everything for you. Compared to models that require complex setups and expensive hardware to run, Ollama significantly lowers the barrier, allowing ordinary users to experience the power of LLMs on their own PCs.

Even better, Ollama will automatically detect your computer's hardware. If you have a powerful GPU, it will prioritize using it for acceleration; if not, it will switch to CPU mode to ensure the model can still run. This "adapt-to-circumstances" smart management allows Ollama to perform its best on various computers.

Moreover, Ollama excels at working with Docker, making the process of deploying and managing large models in containers exceptionally easy, with convenient packaging and migration.

The Charm of Ollama: Why is it so Popular?

Ollama's rapid success is not just due to "local execution"; its advantages are tangible:

• Free and Open, Strong Community: Ollama itself and most of the models it supports are open-source and free. You can use it with confidence, without worrying about your wallet. Moreover, there is an active developer community constantly contributing code, models, and ideas, making Ollama better and better.
• Easy to Get Started, No More Hassle: This point is truly amazing! No complex configuration is needed; installation and running can usually be done with a few commands. For beginners, this is a blessing, saving a lot of time and effort in researching environment configuration.
• Cross-Platform Support, Usable Anywhere: Whether you use Mac, Linux, or Windows, or even Docker, Ollama has installation solutions ready for you. This cross-platform friendliness allows users in different environments to easily get started.
• Rich Model Library, Choose What You Like: From Meta's Llama series to Google's Gemma, to China's Qwen (Tongyi Qianwen), and rising stars like DeepSeek, Ollama's model library has everything (https://ollama.com/library). Want to try one? A single command allows you to download and switch, it couldn't be more convenient!
• Packaging and Management, Well-Organized: Ollama has a mechanism called Modelfile that can package the model's weights, configurations, data, etc., together. This makes model management, customization, and sharing particularly clear and efficient. You can think of it as a model's "recipe file."
• Ability to Call Tools, Transforming into a "Swiss Army Knife": The new version of Ollama already supports models like Llama 3.1 to perform "tool calling." This means the model can, based on your instructions, call external tools it knows (such as search, calculation, etc.) to complete more complex tasks, with huge potential!
• Resource-Friendly, Not Picky: Ollama has put effort into resource optimization, including how to better utilize the GPU. This allows it to run relatively smoothly in resource-constrained environments, without worrying about "burning out" ordinary computers.
• Privacy and Security, Data Localization: Because all calculations are done on your own machine, your data and conversation history are firmly in your control. You don't have to worry about privacy breaches, and it better meets certain data security requirements.

Model Variety: There's Always One for You

Ollama's supported model library continues to grow; you can see the complete list in the official model library (https://ollama.com/library). The community is also constantly contributing new models, making the ecosystem increasingly prosperous.

Here are some of the more popular models to give you a feel (note that model size and parameter count affect the resources required to run):

Model Name	Features/Version	Parameters	Size (approx.)	Quick Run Command (ollama run ...)
DeepSeek-R1	-	7B	4.7GB	deepseek-r1
DeepSeek-R1	-	671B	404GB	deepseek-r1:671b
Llama 3.3	-	70B	43GB	llama3.3
Llama 3.2	-	3B	2.0GB	llama3.2
Llama 3.2	-	1B	1.3GB	llama3.2:1b
Llama 3.2 Vision	Vision Model	11B	7.9GB	llama3.2-vision
Llama 3.2 Vision	Vision Model	90B	55GB	llama3.2-vision:90b
Llama 3.1	-	8B	4.7GB	llama3.1
Llama 3.1	-	405B	231GB	llama3.1:405b
Gemma 2	-	2B	1.6GB	gemma2:2b
Gemma 2	-	9B	5.5GB	gemma2
Gemma 2	-	27B	16GB	gemma2:27b
Mistral	-	7B	4.1GB	mistral:7b
Qwen	-	110B	63GB	qwen:110b
Phi 4	-	14B	9.1GB	phi4
CodeLlama	Code Generation	70B	39GB	codellama:70b
Qwen2	-	72B	41GB	qwen2:72b
Llava	Vision Model	7B	4.7GB	llava:7b
Nomic Embed Text	Text Embedding Model	v1.5	274MB	nomic-embed-text:v1.5 (use pull command to download)

Ollama is like a large model "local launcher" and "manager," greatly reducing the barrier to experiencing and using cutting-edge AI technology on personal computers. If you are interested in large models and want to try them out, then Ollama is definitely worth learning about and trying!