ThomasBaruzier's picture
Upload perplexity.md
449ed48 verified
|
raw
history blame
879 Bytes
Qwen2.5-1.5B-Instruct
Quant Size (MB) PPL Size (%) Accuracy (%) PPL error rate
IQ1_S 417 193.6245 1.77149
IQ1_M 443 66.9068 0.52878
IQ2_XXS 488 33.3356 0.25559
IQ2_XS 525 20.2870 0.14936
IQ2_S 538 18.2927 0.13380
IQ2_M 574 15.4838 0.11113
Q2_K_S 611 16.0169 0.11623
IQ3_XXS 638 12.3935 0.08770
Q2_K 645 14.1657 0.10105
IQ3_XS 698 11.7112 0.08256
Q3_K_S 726 12.4782 0.08842
IQ3_S 728 11.4241 0.07977
IQ3_M 741 11.4058 0.07862
Q3_K_M 786 11.3529 0.08018
Q3_K_L 840 11.1934 0.07913
IQ4_XS 855 10.5302 0.07351
IQ4_NL 893 10.5116 0.07335
Q4_0 895 10.8217 0.07576
Q4_K_S 897 10.5236 0.07360
Q4_K_M 941 10.4628 0.07310
Q4_1 970 10.5100 0.07347
Q5_K_S 1048 10.2715 0.07148
Q5_0 1051 10.3196 0.07212
Q5_K_M 1073 10.2529 0.07143
Q5_1 1126 10.2624 0.07140
Q6_K 1214 10.2030 0.07108
Q8_0 1571 10.1670 0.07068
F16 2951 10.1512 0.07058