Table 10.

Comparison of the perplexity and accuracy of our compression method with those of QuantGPT, KnGPT2, LightPAFF, and SparseGPT.

Dataset	CB		MR		HS
Compression method	PPL (↓)	Accuracy ($\uparrow$)	PPL (↓)	Accuracy ($\uparrow$)	PPL (↓)	Accuracy ($\uparrow$)
QuantGPT 17	45.58	88.13	55.25	80.07	45.93	78.85
KnGPT2 18	23.28	89.25	53.73	81.27	56.8	75.41
LightPAFF 19	20.50	92.18	38.78	85.17	37.50	85.47
SparseGPT57	16.14	92.65	23.88	98.13	30.43	88.36
Our method	14.28	93,72	20.14	91,86	26.84	90.13

Significant values are in bold.