Qwen3-0.6B

pack Qwen3-0.6B@282f68b573fc832a (id 282f68b573fc832a), 2 result(s). Lower KLD is better. Click a column to sort.

KLD mean against file size. Bars are 95% intervals. Lower and further left is better.

Quant	Publisher	Size	KLD mean	95% interval	Worst 1%	Top-1	PPL ratio	Loops (q/ref)	Wrong script	JSON ok	Backend
Q3_K_M	tensorblock	414.0 MB	0.3953	0.3803 to 0.4104	3.017	69.8%	1.455	35/32	0	2/8	llamacpp
Q2_K	tensorblock	347.3 MB	2.4554	2.3736 to 2.5431	10.484	33.8%	11.019	48/32	0	0/8	llamacpp

approx = the server only reported its own top-N, so some reference tokens were counted in the tail bucket. nd = the same text gave slightly different numbers on two runs. "Loops" shows quant / original. All results used the same probe set and pack.