data/tiny_corpus.txt

The little machine listened to the rain and learned one letter at a time.
A thought began as a spark, then became a pattern, then became a voice.
The teacher said: do not fear small beginnings. A tiny model can still teach us the shape of learning.

In the quiet workshop, tokens walked in a line.
Each token looked back at the tokens before it and asked, what should come next?
The answer was never magic. It was counting, guessing, correcting, and trying again.

A fox told a crow: wisdom is not the size of the library, but the care of the attention.
A crow replied: even a small mind can remember a melody if it hears the song often enough.
The river laughed, because the river had trained every stone by repeating its lesson.

We build tinyllm to understand language models from the inside.
We train on CPU because patience is part of the experiment.
We start with characters because characters are honest: small marks, simple rules, many possibilities.

The model begins with random noise.
Then loss falls a little.
Then letters become syllables.
Then syllables become words.
Then words begin to imitate the books that fed them.

This is not a giant assistant.
This is a seed.
A seed does not pretend to be a forest.
A seed shows that a forest is possible.

The student asked: can a tiny model think?
The teacher answered: first let it predict. Then let us study what prediction teaches.

Again the little machine listened.
Again the optimizer stepped.
Again the text became less strange.
And in the warm hum of the CPU, the tiny language model began.