Name: AI Scaling Trends Lab
Author: MySimulator

Model configuration

Parameters: 7B Training tokens: 140B Cost per GPU-hour: $2.50

Parameters and tokens are shown on a log2 slider (8 → 256, spanning roughly 256M to a few hundred billion) since real model sizes span many orders of magnitude.

Estimates

Training FLOPs—

Estimated training cost—

Tokens per parameter—

Chinchilla-optimal ratio?—

Estimated loss vs. training compute (this configuration marked)