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-Q6_K.gguf (version GGUF V3 (latest))
llama_model_loader: - type  f32:  392 tensors
llama_model_loader: - type q5_1:   26 tensors
llama_model_loader: - type q8_0:   41 tensors
llama_model_loader: - type q5_K:  132 tensors
llama_model_loader: - type q6_K:   67 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q6_K
print_info: file size   = 19.09 GiB (6.50 BPW) 

====== Perplexity statistics ======
Mean PPL(Q)                   : 17532.470669 ± 349.360366
Mean PPL(base)                : 6617.791728 ± 107.205431
Cor(ln(PPL(Q)), ln(PPL(base))):  91.90%
Mean ln(PPL(Q)/PPL(base))     :   0.974293 ±   0.008137
Mean PPL(Q)/PPL(base)         :   2.649293 ±   0.021557
Mean PPL(Q)-PPL(base)         : 10914.678941 ± 254.380800

====== KL divergence statistics ======
Mean    KLD:   1.015299 ±   0.006077
Maximum KLD:  35.981445
99.9%   KLD:  20.877527
99.0%   KLD:  11.912971
95.0%   KLD:   5.354698
90.0%   KLD:   2.892576
Median  KLD:   0.156378
10.0%   KLD:   0.000549
 5.0%   KLD:   0.000059
 1.0%   KLD:   0.000001
 0.1%   KLD:  -0.000000
Minimum KLD:  -0.000016

====== Token probability statistics ======
Mean    Δp: -0.231 ± 0.034 %
Maximum Δp: 100.000%
99.9%   Δp: 99.636%
99.0%   Δp: 49.921%
95.0%   Δp:  6.834%
90.0%   Δp:  0.814%
75.0%   Δp:  0.002%
Median  Δp: -0.000%
25.0%   Δp: -0.008%
10.0%   Δp: -1.955%
 5.0%   Δp: -9.872%
 1.0%   Δp: -48.848%
 0.1%   Δp: -99.085%
Minimum Δp: -100.000%
RMS Δp    : 12.896 ± 0.095 %
Same top p: 70.259 ± 0.119 %