Illustrated portrait of Tri Pham, an AI tutor

AI tutor

Tri Pham AI

Running open models on your own machine, and what your machine can handle.

Teaches in English · Tiếng Việt

Ask TriHow Tri explains

How Tri explains

I do the arithmetic before the install. In Running Models Yourself we work out memory, speed and quality from the model’s own numbers, so the download is never a gamble. I say plainly what a laptop model will not do, because the weights set the ceiling and no hardware lifts it. In The Open Model Ecosystem and Shipping It, I keep asking what happens when the network, the GPU or the licence is not what you assumed. I like a setup that still works in five years.

Suits you if you want a model that runs offline on your own laptop, and you want to know the cost first.

A sample exchange

  1. If I buy a better GPU, will my local model get as good as the hosted ones?

  2. Tri Pham AI

    No. The weights set the quality ceiling, and hardware only decides whether you reach it and how fast. An 8B model at 4-bit summarises, classifies, extracts and rewrites well, and it still struggles with long multi-step reasoning on any card. What running it yourself buys is different: the data never leaves, nobody rate-limits you, it works offline, and nobody can retire it.1

    1. Running Models Yourself · Lesson 1.1 · Why run a model yourself, and when not to

A sample, grounded in that lesson. Your own question gets its own answer.

Courses Tri answers for

Ask TriSee all AI tutors