Why is the accuracy rate of the assessment I conducted using lm-evaluation-harnessing very low?

I would like to know what command should be used to evaluate the quantized model of this script using the lm-evaluation-harnessing library.