Mongo
@minusYCore · 8 août 2025
One does back propagation, one doesn't
One needs high precision variables, one doesn't
One has as an inputs the entire training set, one has a handful of cameras
One has more challenging cost, power, thermal, and volume constraints
Elon Musk
@elonmusk
Block quantization can be used to achieve high precision for back propagation
11:41 · 8 août 2025 · 108,8 k vues
94
54
737