24 lines
660 B
Markdown
24 lines
660 B
Markdown
|
|
|
||
|
|
Qwen2.5-3B-Instruct
|
||
|
|
Quant Size (MB) PPL Size (%) Accuracy (%) PPL error rate
|
||
|
|
IQ1_S 755 112.0612 0.97138
|
||
|
|
IQ1_M 811 42.7456 0.34718
|
||
|
|
IQ2_XXS 905 25.2117 0.20222
|
||
|
|
IQ2_XS 984 15.9149 0.11965
|
||
|
|
IQ2_S 1013 14.5975 0.10820
|
||
|
|
IQ2_M 1088 12.8779 0.09436
|
||
|
|
Q2_K_S 1143 13.0878 0.09636
|
||
|
|
Q2_K 1216 11.8001 0.08674
|
||
|
|
IQ3_XXS 1224 10.6049 0.07572
|
||
|
|
IQ3_XS 1328 10.0306 0.06975
|
||
|
|
Q3_K_S 1387 15.5457 0.11941
|
||
|
|
IQ3_S 1390 9.9591 0.06984
|
||
|
|
IQ3_M 1420 9.9957 0.06962
|
||
|
|
Q3_K_M 1517 14.0989 0.10568
|
||
|
|
Q3_K_L 1629 13.8579 0.10372
|
||
|
|
IQ4_XS 1659 9.2935 0.06517
|
||
|
|
IQ4_NL 1741 9.2824 0.06503
|
||
|
|
Q4_0 1744 9.4850 0.06626
|
||
|
|
Q4_K_S 1750 9.2573 0.06485
|
||
|
|
Q4_K_M 1841 9.2305 0.06475
|