Quantization documentation runnable in Colab

[jchwenger]

Is your feature request related to a problem? Please describe.

Hi there, I was glad to find this Quantization example in your Quickstart, promising to run Qwen-Image using a little under 15GB, which sounds like it could just about work in Colab Free Tier (for students). Sadly, nope, OOM. Given the number, I'm assuming you had Colab in mind, or no?

Describe the solution you'd like.

I know it's quite a big ask, since the free infrastructure like free GPUs on Colab are lagging quite far behind the usual memory requirements for Diffusion/LLMs etc., but for teaching purposes it would real nice to still have examples that can be run without having to pay for a subscription (again having students in mind).

Describe alternatives you've considered.

I'm trying to find either quantization strategies, or equivalent recent diffusion models to the lovely tiny LLMs (< 1B params) that have recently surfaced from various labs, to be run on free T4 or even on personal machines, so that it's easy enough for students to play with them easily.

I'll check other models, but I was wondering if the Quickstart example was meant to be run on a T4, or in fact not.