ThomasBaruzier's picture
Upload perplexity.md
cc0571a verified
|
raw
history blame contribute delete
No virus
875 Bytes

Llama-3.2-1B-Instruct Quant Size (MB) PPL Size (%) Accuracy (%) PPL error rate IQ1_S 376 771.8958 14.99148 IQ1_M 395 162.0038 2.86547 IQ2_XXS 427 46.0426 0.77657 IQ2_XS 454 30.7626 0.50736 IQ2_S 467 25.4944 0.41940 IQ2_M 492 21.1112 0.34245 Q2_K_S 529 24.5117 0.40072 IQ3_XXS 537 17.2479 0.27837 Q2_K 554 26.1688 0.44789 IQ3_XS 593 16.0104 0.25685 Q3_K_S 612 19.1038 0.31660 IQ3_S 615 15.6453 0.24806 IQ3_M 627 15.4512 0.24445 Q3_K_M 659 14.9000 0.23958 Q3_K_L 699 14.7286 0.23679 IQ4_XS 709 14.1783 0.22704 IQ4_NL 738 14.1777 0.22727 Q4_0 738 14.4071 0.23021 Q4_K_S 740 14.0726 0.22511 Q4_K_M 771 14.0496 0.22523 Q4_1 794 14.1039 0.22552 Q5_K_S 852 13.8515 0.22187 Q5_0 854 13.8766 0.22210 Q5_K_M 870 13.8295 0.22162 Q5_1 910 13.7981 0.22042 Q6_K 975 13.7604 0.22054 Q8_0 1260 13.7166 0.21964 F16 2365 13.7126 0.21966