Neural Network Internals

Quantisation
Precision Explorer

Adjust a weight value and watch how FP32, INT8, and INT4 representations diverge — with live memory and error statistics.

Weight Value 0.318

−2.0000+2.000

Model Parameters 7B

1B35B70B

Memory Footprint 7B parameter model