AI Trainer

Paste train and test losses. Detect grokking — the phase transition from memorization to understanding. 224,000× cheaper than memorization, measured in Landauer bits.

LINE 1: TRAIN LOSSES LINE 2: TEST LOSSES (comma-separated, one line each)

Loading runtime...

Line 1: train losses. Line 2: test losses. Press Run.

Everything is free. If this helped → support GUMP

FOUR PHASES

LEARNING — both losses high, model hasn’t started. MEMORIZED — train loss low but test still high, gap widening. GROKKING — test loss suddenly drops, gap closing. The phase transition. UNDERSTOOD — both low, gap stable. Understanding achieved.

INSTALL

pip install begump
from gump.aitrainer import detect_phase

phase = detect_phase(train_losses, test_losses)
print(phase)

Your data stays on your machine.

GUMP — Tools · Support · [email protected]