Loading slide

What a Trained Model Actually Is

  1. 01What a Trained Model Actually Is
  2. 02Memory card
  3. 03A model is a file
  4. 04Two files
  5. 05How the two fit together
  6. 06A Piano
  7. 07The valuable file
  8. 08You connect to it
  9. 09Where does training data come from?
  10. 10Cleaning the data
  11. 11The network can't read
  12. 12Guess the next word
  13. 13Right or wrong
  14. 14One nudge, a billion times
  15. 15Take a breath
  16. 16Where your memories live
  17. 17A base model is not an assistant
  18. 18Why it's called pre-training
  19. 19The weights are the model
  20. 20The scale of it
  21. 21Why so many?
  22. 22What a weight actually is
  23. 23Everywhere and nowhere
  24. 24Using the model: inference
  25. 25The whole arrangement
  26. 26Reinforce your understanding
  27. 27Question: What is a model, really?
  28. 28Question: Learning from raw text
  29. 29Question: The model doesn't know what it doesn't know
  30. 30Quiz: answer
  31. 31The machine learned something real.
  32. 32Want to go deeper?
7 / 31
BackNext

The valuable file

The weights file is what took the effort to produce. Months of computation, enormous amounts of electricity, hundreds of millions of dollars, all spent adjusting billions of numbers until they work. The program file is public knowledge. Researchers publish it openly. The weights are what companies guard, sell access to, and spend fortunes producing.

Some companies keep their weights entirely private. You can use the model through their website or app, but the file itself never leaves their servers. Others release their weights openly, making them available for anyone to download, study, or build on.

This distinction matters. An open weights model can run on your own hardware, be inspected by researchers, or be adapted for specific uses. A closed model can only be used on the terms the company sets.

Think about eating a dish at a restaurant that you love. You can see what's in it. You can taste the ingredients. You could probably name most of them. But you couldn't go home and make it. Not really. You'd need the exact quantities, the order things were added, the temperature, the timing. All the knowledge that makes it that dish rather than a rough approximation of it.

Knowing the program is a bit like knowing the ingredients. You can see the structure. But the weights, the specific configuration that took months and hundreds of millions of dollars to produce, that's the recipe. And most companies keep it locked in the kitchen.

It's why what separates one model from another, what makes different from different from , is almost entirely in the weights. The pianist, not the piano.

The kind of model we've been building toward has a specific name: a Large Language Model, or LLM. Large, because of the sheer number of weights. Language, because it was trained on text. Model, because it is exactly what we've been describing: a file full of numbers.

But where does that file actually live, and how do you get to it?

Citations