microgpt-rs

Character-level GPT trained on names, running in WebAssembly. Based on Andrej Karpathy's microgpt (100k steps instead of 1k). Source code.

Dataset: Using default (32,033 names)

Sample names during training

Generated names

What comes after ___?

How temperature affects these predictions

T = 0.2
Conservative — always picks the obvious choice
T = 0.8
Balanced — natural variety
T = 1.8
Creative — wild, unexpected picks

How surprising is this name?