Incognition logo
Home
Learn
Practice
Leaderboard
0
VisualizationsGenerative & Applied Deep Learning

Quantization

Pick 32, 16, 8 or 4 bits per weight and watch a 7-billion-weight model shrink from 28 GB to 3.5 GB. Fewer bits, smaller model.

7B model size
28.0 GB

Weights are just numbers, and nothing says they need 32 bits each. Use 8 and the model is four times smaller.

Customize