ThomasBaruzier commited on
Commit
cc0571a
1 Parent(s): 8ad119f

Upload perplexity.md

Browse files
Files changed (1) hide show
  1. perplexity.md +30 -0
perplexity.md ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Llama-3.2-1B-Instruct
2
+ Quant Size (MB) PPL Size (%) Accuracy (%) PPL error rate
3
+ IQ1_S 376 771.8958 14.99148
4
+ IQ1_M 395 162.0038 2.86547
5
+ IQ2_XXS 427 46.0426 0.77657
6
+ IQ2_XS 454 30.7626 0.50736
7
+ IQ2_S 467 25.4944 0.41940
8
+ IQ2_M 492 21.1112 0.34245
9
+ Q2_K_S 529 24.5117 0.40072
10
+ IQ3_XXS 537 17.2479 0.27837
11
+ Q2_K 554 26.1688 0.44789
12
+ IQ3_XS 593 16.0104 0.25685
13
+ Q3_K_S 612 19.1038 0.31660
14
+ IQ3_S 615 15.6453 0.24806
15
+ IQ3_M 627 15.4512 0.24445
16
+ Q3_K_M 659 14.9000 0.23958
17
+ Q3_K_L 699 14.7286 0.23679
18
+ IQ4_XS 709 14.1783 0.22704
19
+ IQ4_NL 738 14.1777 0.22727
20
+ Q4_0 738 14.4071 0.23021
21
+ Q4_K_S 740 14.0726 0.22511
22
+ Q4_K_M 771 14.0496 0.22523
23
+ Q4_1 794 14.1039 0.22552
24
+ Q5_K_S 852 13.8515 0.22187
25
+ Q5_0 854 13.8766 0.22210
26
+ Q5_K_M 870 13.8295 0.22162
27
+ Q5_1 910 13.7981 0.22042
28
+ Q6_K 975 13.7604 0.22054
29
+ Q8_0 1260 13.7166 0.21964
30
+ F16 2365 13.7126 0.21966