A Karpathy-style, minimal, pedagogical implementation of a transformer that does in-context learning on tabular data. ~500 lines of Python. No pretrained weights needed. Runs on CPU in minutes. No ...