Quantisation workbench

Compress real numeric weights into INT8 or INT4 codes and measure the reconstruction error.

How this experiment works

Symmetric per-tensor quantisation uses a scale based on the largest absolute weight, then rounds and reconstructs each value.

What it does not prove

This is a mathematical weight demonstration. It does not measure an actual model’s speed, accuracy or deployed memory use.

Try this challenge

Add a large outlier. Watch what happens to the smaller weights.

Related Trust Agent courses · Sources and licences

Enable JavaScript to run the experiment on your device.