Forward-Forward Binary Language Model

Hinton's Forward-Forward algorithm (no backprop, no GPU) learning next-token (byte / bigram / trigram) prediction via a bit readout. Train, pause, predict, resume.

Model & Training

3
128
8
0.020
32
Parameters (weights)–
Optimizer state–
Total training RAM–

Training stats

Statusidle
Examples seen0
Steps/sec0
Next-byte bit / exact acc–
Mean goodness +/−–
Live sample

Predict / Generate (uses current weights)

0.80
120
Drop a text file to use as training data