ThomasBaruzier commited on
Commit
c5f48d3
1 Parent(s): bd8b910

Upload perplexity.md

Browse files
Files changed (1) hide show
  1. perplexity.md +23 -0
perplexity.md ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Llama-3.2-3B-Instruct
2
+ Quant Size (MB) PPL Size (%) Accuracy (%) PPL error rate
3
+ IQ1_S 828 125.0970 1.99765
4
+ IQ1_M 882 51.1917 0.82201
5
+ IQ2_XXS 971 24.6228 0.37767
6
+ IQ2_XS 1050 17.6591 0.27116
7
+ IQ2_S 1101 15.8955 0.24655
8
+ IQ2_M 1173 14.5399 0.22581
9
+ Q2_K_S 1216 15.7948 0.24709
10
+ IQ3_XXS 1287 12.7005 0.19429
11
+ Q2_K 1301 14.8843 0.23696
12
+ IQ3_XS 1409 12.5168 0.19188
13
+ IQ3_S 1472 12.2121 0.18863
14
+ Q3_K_S 1472 12.8759 0.20140
15
+ IQ3_M 1526 11.8347 0.18147
16
+ Q3_K_M 1610 11.6367 0.18088
17
+ Q3_K_L 1732 11.5900 0.18091
18
+ IQ4_XS 1745 11.3192 0.17504
19
+ IQ4_NL 1829 11.3142 0.17506
20
+ Q4_0 1833 11.3154 0.17484
21
+ Q4_K_S 1839 11.2630 0.17415
22
+ Q4_K_M 1926 11.2436 0.17406
23
+ Q4_1 1997 11.2838 0.17446