ThomasBaruzier's picture
Upload perplexity.md
c5f48d3 verified
|
raw
history blame
No virus
697 Bytes
Llama-3.2-3B-Instruct
Quant Size (MB) PPL Size (%) Accuracy (%) PPL error rate
IQ1_S 828 125.0970 1.99765
IQ1_M 882 51.1917 0.82201
IQ2_XXS 971 24.6228 0.37767
IQ2_XS 1050 17.6591 0.27116
IQ2_S 1101 15.8955 0.24655
IQ2_M 1173 14.5399 0.22581
Q2_K_S 1216 15.7948 0.24709
IQ3_XXS 1287 12.7005 0.19429
Q2_K 1301 14.8843 0.23696
IQ3_XS 1409 12.5168 0.19188
IQ3_S 1472 12.2121 0.18863
Q3_K_S 1472 12.8759 0.20140
IQ3_M 1526 11.8347 0.18147
Q3_K_M 1610 11.6367 0.18088
Q3_K_L 1732 11.5900 0.18091
IQ4_XS 1745 11.3192 0.17504
IQ4_NL 1829 11.3142 0.17506
Q4_0 1833 11.3154 0.17484
Q4_K_S 1839 11.2630 0.17415
Q4_K_M 1926 11.2436 0.17406
Q4_1 1997 11.2838 0.17446