The NanoChat Speedrun · Part 2
NanoChat Speedrun, Part 2: What the Run Actually Printed
Eight predictions, one H100 node, and a log file that answered most of them in the first forty lines.
nanochat gpu training llm
Anand Nair
Sufficiently advanced technology, taken seriously.
The NanoChat Speedrun · Part 1
NanoChat Speedrun, Part 1: The Code That Makes H100s Go Brrrr
What I expected before spending anything on GPUs — a close reading of NanoChat's training stack, and the predictions it let me make.
nanochat gpu training llm