ThomasBaruzier's picture
Upload perplexity.md
151493b verified
Quant Size (MB) PPL Size (%) Accuracy (%) PPL error rate
IQ1_S 489 355.5202 14.40 9.80 8.19
IQ1_M 515 159.7892 15.17 21.80 3.59
IQ2_XXS 560 113.7759 16.49 30.62 2.62
IQ2_XS 597 66.9608 17.58 52.03 1.44
IQ2_S 633 55.5589 18.65 62.70 1.18
IQ2_M 668 49.2839 19.67 70.69 1.04
Q2_K_S 683 56.3132 20.12 61.86 1.22
Q2_K 718 52.9062 21.15 65.85 1.15
IQ3_XXS 733 40.4841 21.59 86.05 0.85
IQ3_XS 793 37.9592 23.36 91.78 0.78
Q3_K_S 821 40.8887 24.18 85.20 0.86
IQ3_S 822 37.3800 24.21 93.20 0.77
IQ3_M 836 35.4048 24.62 98.40 0.72
Q3_K_M 881 36.7423 25.95 94.82 0.77
Q3_K_L 935 37.9153 27.54 91.88 0.80
IQ4_XS 972 37.3032 28.63 93.39 0.79
IQ4_NL 1018 37.4248 29.99 93.09 0.79
Q4_0 1019 36.6871 30.01 94.96 0.77
Q4_K_S 1021 35.4185 30.07 98.36 0.74
Q4_K_M 1065 35.2520 31.37 98.82 0.73
Q4_1 1108 35.9411 32.64 96.93 0.75
Q5_K_S 1200 35.3946 35.35 98.43 0.74
Q5_0 1203 36.2193 35.43 96.18 0.76
Q5_K_M 1225 34.7903 36.08 100.14 0.72
Q5_1 1292 35.1618 38.06 99.08 0.73
Q6_K 1396 34.7554 41.12 100.24 0.72
Q8_0 1806 34.9765 53.19 99.60 0.73
F16 3395 34.8374 100 100 0.73