# The idea behind it — M.E.F LLM Studio

Source: https://www.mef-llm-studio.com/en/story/

## Why M.E.F LLM Studio came about

By Mike Walter

Developed by Mike Walter, the developer behind SmartFTP.
[Mentioned in 2005](https://www.patheos.com/blogs/drishtikone/2005/11/chat-room-changed-world/)

Working with LLMs raised more and more questions. Eventually, I wanted to look closer myself and find a way to explain it to others.

### Where is the magic?

The more I worked with language models, the more often I wondered: how does this actually work? I type something and get an answer. Sometimes it fits surprisingly well; sometimes it gets things wrong. But what happens between my input and that answer?

Using the models was one thing. Understanding how text becomes numbers, what a model learns during training and how it produces an output step by step was another. Each answer brought more questions.

### How do I explain this to my family?

It was not just about my own curiosity. How would I explain this to my son or my wife? Both use LLMs, but using one does not mean knowing what it actually is or how it works.

Terms like token, attention or training are not much help if you cannot picture them. I wanted a way for us to look at it together: here is our text. This is how it is split up. The model computes possible continuations. A token is selected and appended, and the process repeats.

### Trying it, not just reading about it

That became the idea for M.E.F LLM Studio: an application where you can try these steps yourself. Use your own text, build a tokenizer, train a small model and then explore what it generates.

Small models are interesting for this because you can experiment with them. What changes with different training data? What does temperature do? Why is the most likely token not always selected? The explanations should connect to something you have just observed for yourself.

That is why visualisation is central to the idea. A good-looking display alone is not enough for me. It should help you understand a step or ask a question that might not otherwise have occurred to you.

### Making the unknown a little less intimidating

For me, the word “magic” also carries some unease. When a system answers so convincingly and you do not understand how, it is hard to judge. What can I rely on it for? When should I ask more questions or check something?

I want to take away a little of that fear by making the process more approachable. That does not make everything simple, and a small experiment will not explain every behaviour of a large model. But having an understandable way in makes a difference: you no longer have to simply marvel at the output.

### The questions continue

M.E.F LLM Studio grew out of these questions. It is a learning project for me, too. I hope others can use it to find their own way in, whether out of curiosity, in a classroom or at home, exploring together what is behind the apparent magic.


Contact: support@mef-llm-studio.com