ThomasBaruzier's picture
Upload perplexity.md
43933b7 verified
|
raw
history blame
871 Bytes
Qwen2.5-0.5B-Instruct
Quant Size (MB) PPL Size (%) Accuracy (%) PPL error rate
IQ1_S 302 30.0453 0.23714
IQ1_M 304 24.9151 0.19238
IQ2_XXS 307 21.4864 0.16704
IQ2_XS 310 19.7829 0.15355
IQ2_S 311 19.2041 0.14841
IQ2_M 314 18.3250 0.14001
Q2_K_S 316 18.3462 0.14091
IQ3_XXS 319 16.9784 0.12828
Q3_K_S 323 17.7765 0.13668
IQ3_S 323 16.3794 0.12173
IQ3_XS 323 16.3794 0.12173
Q2_K 323 17.6841 0.13561
IQ3_M 327 16.3667 0.12182
IQ4_XS 334 15.8792 0.11933
IQ4_NL 337 15.8468 0.11921
Q4_0 337 17.1007 0.13053
Q3_K_M 339 15.8499 0.11934
Q3_K_L 353 15.7298 0.11820
Q4_1 358 16.1819 0.12328
Q4_K_S 368 15.5497 0.11716
Q5_0 380 15.5038 0.11702
Q4_K_M 380 15.4428 0.11637
Q5_K_S 394 15.5266 0.11682
Q5_1 400 15.4875 0.11641
Q5_K_M 401 15.4788 0.11631
Q6_K 483 15.2145 0.11422
Q8_0 507 15.2390 0.11452
F16 949 15.1819 0.11400