About

I'm Prashanth. This is my notebook for learning how large language models run in production: how a single token gets generated, why speed is limited by memory, and how serving systems like vLLM squeeze more out of a GPU.

Every concept page tries to do three things:

• Explain the idea in plain language first, math second.
• Give you a playground to change the numbers and see what happens.
• Draw the mechanism, so it's something you can picture.

The path is the plan I'm following, and the log is where I write down what I did each week. I'm learning in public, so expect mistakes. If you spot one, I'd like to hear about it.

My main site is basani.me.