Ollama Model Family Tree for GitHub Browsers [Archived Edition]

The AI scene has seen too many models lately.

When you open the Ollama library, you see Qwen, Gemma, Llama, DeepSeek, Mistral, Phi...

To a newcomer, they might all look the same. However, when you browse GitHub, you start to see a slightly different landscape.

(´･ω･`) For those who don't have a high-spec PC, what you want isn't the "strongest model." You are looking for a partner to help you understand interesting repositories found on GitHub, save them as Markdown, and turn them into your own knowledge base.

So, this time, rather than a performance ranking, I would like to organize them as a family tree based on "what the model was born to do."

The general term for small AI models that run in local environments like Ollama is "SLM (Small Language Models)."

It is a standard term used to refer to models designed to operate efficiently even in resource-constrained environments, as a counterpart to large-scale "LLMs (Large Language Models)."

First Force: The Versatile Qwen Family

The family that currently feels the most momentum.

They can converse. They can read code. They are good at summarizing.
They are also relatively strong in Japanese.

For GitHub browsers, they are becoming the go-to choice: "When in doubt, use Qwen." Qwen models, in particular, handle general chores such as:

with ease.

To use a Unix analogy, they are like "just use bash."

Second Force: Google's Gemma Family

If I had to describe Gemma in one word, it would be "smart small model."

They do not assume the presence of a massive GPU.

Because they run even in relatively small configurations, they are suitable for:

It has good compatibility with.

For GitHub browsing purposes, it demonstrates sufficient capability in reading and summarizing READMEs.

Personally, I have a strong impression of it as a 'humanities-oriented model to keep on hand'.

Third Force: The Royal Llama Family

Llama is the main branch in terms of the family tree.

Current local LLM culture is heavily influenced by Llama in many ways.

However, recently, Qwen and Gemma have grown rapidly, so it is no longer the first model beginners choose. Even so,

If you trace the ancestors of many derivative models, you end up at Llama.

When browsing GitHub, there are still many situations where you think, 'Is this model a Llama derivative after all?' It is an existence you cannot avoid if you want to learn history.

Fourth Force: The Reasoning-Specialized DeepSeek Family

DeepSeek is a bit different.

Its feature is 'thinking'. It reasons before answering.

In human terms,

it is not a person who answers reflexively, but a person whowrites notes in a notebook before
answering.

It is interesting for GitHub code explanations and design understanding.

However, in terms of lightness, Qwen and Gemma have the advantage.

Fifth Force: The Artisan Group of the Coder Family

This is more of a professional group than an independent family line. Representatives are,

and so on. Their job is clear: reading code.

If you're a GitHub browser, you should try this at least once. More than the README,

they have a strength in understanding actual source code.

The future I see

What has been increasing on GitHub lately is not "people who use giant models"

but "people who have multiple small models share roles".

For example,

GitHub　↓
Crawl4AI　↓
Markdown化　↓
Gemmaで要約　↓　Qwenで整理　↓　
Logseq、Memos保存、（Obsidianも可）

is the kind of flow.

This is not just a game for people with massive GPUs.

It can be fully enjoyed on a regular PC with 16GB or 32GB of memory. By the way, I have 8GB and (´･ω･`) sigh

Summary

There are many Ollama models. However, from the perspective of a GitHub browser,

it is actually not that complicated.

For starters, just remembering this map is enough. And what is important is,

not collecting models.

It is being able to save the knowledge you find on GitHub in your own words.

Local LLMs are merely new tools for that purpose.

#Ollama #LocalLLM #Qwen #Gemma
#GitHub #AIUtilization