llama_model_loader: loaded meta data with 54 key-value pairs and 658 tensors from gemma-4-26B-A4B-it-WIP/gemma-4-26B-A4B-it-Q3_K.gguf (version GGUF V3 (latest))
llama_model_loader: - type  f32:  392 tensors
llama_model_loader: - type q2_K:   33 tensors
llama_model_loader: - type q3_K:    2 tensors
llama_model_loader: - type q4_K:    2 tensors
llama_model_loader: - type iq3_xxs:  124 tensors
llama_model_loader: - type iq4_nl:   60 tensors
llama_model_loader: - type iq3_s:    8 tensors
llama_model_loader: - type iq4_xs:   37 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = IQ4_XS - 4.25 bpw
print_info: file size   = 10.28 GiB (3.50 BPW) 

====== Perplexity statistics ======
Mean PPL(Q)                   : 8598.460723 ± 162.541607
Mean PPL(base)                : 6617.791728 ± 107.205431
Cor(ln(PPL(Q)), ln(PPL(base))):  80.01%
Mean ln(PPL(Q)/PPL(base))     :   0.261821 ±   0.011391
Mean PPL(Q)/PPL(base)         :   1.299295 ±   0.014800
Mean PPL(Q)-PPL(base)         : 1980.668995 ± 100.145541

====== KL divergence statistics ======
Mean    KLD:   3.945886 ±   0.012077
Maximum KLD:  42.292805
99.9%   KLD:  28.313078
99.0%   KLD:  19.872719
95.0%   KLD:  13.435264
90.0%   KLD:  10.394227
Median  KLD:   2.267470
10.0%   KLD:   0.016854
 5.0%   KLD:   0.002028
 1.0%   KLD:   0.000047
 0.1%   KLD:   0.000001
Minimum KLD:  -0.000001

====== Token probability statistics ======
Mean    Δp: -0.107 ± 0.063 %
Maximum Δp: 100.000%
99.9%   Δp: 99.966%
99.0%   Δp: 94.442%
95.0%   Δp: 34.712%
90.0%   Δp:  6.938%
75.0%   Δp:  0.031%
Median  Δp: -0.000%
25.0%   Δp: -0.027%
10.0%   Δp: -7.778%
 5.0%   Δp: -36.028%
 1.0%   Δp: -93.667%
 0.1%   Δp: -99.960%
Minimum Δp: -100.000%
RMS Δp    : 23.984 ± 0.101 %
Same top p: 41.573 ± 0.129 %