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-Q4_K.gguf (version GGUF V3 (latest))
llama_model_loader: - type  f32:  392 tensors
llama_model_loader: - type q4_1:   22 tensors
llama_model_loader: - type q5_0:   31 tensors
llama_model_loader: - type q3_K:    9 tensors
llama_model_loader: - type q4_K:   59 tensors
llama_model_loader: - type q5_K:    1 tensors
llama_model_loader: - type iq4_nl:    7 tensors
llama_model_loader: - type iq4_xs:  137 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q4_K - Medium
print_info: file size   = 13.22 GiB (4.50 BPW) 

====== Perplexity statistics ======
Mean PPL(Q)                   : 19441.390960 ± 385.285232
Mean PPL(base)                : 6617.791728 ± 107.205431
Cor(ln(PPL(Q)), ln(PPL(base))):  87.72%
Mean ln(PPL(Q)/PPL(base))     :   1.077643 ±   0.009589
Mean PPL(Q)/PPL(base)         :   2.937746 ±   0.028169
Mean PPL(Q)-PPL(base)         : 12823.599232 ± 295.758922

====== KL divergence statistics ======
Mean    KLD:   2.067631 ±   0.008670
Maximum KLD:  38.416069
99.9%   KLD:  24.495975
99.0%   KLD:  15.661798
95.0%   KLD:   9.053732
90.0%   KLD:   6.103799
Median  KLD:   0.685425
10.0%   KLD:   0.003049
 5.0%   KLD:   0.000337
 1.0%   KLD:   0.000006
 0.1%   KLD:   0.000000
Minimum KLD:  -0.000004

====== Token probability statistics ======
Mean    Δp: -0.855 ± 0.046 %
Maximum Δp: 100.000%
99.9%   Δp: 99.864%
99.0%   Δp: 69.856%
95.0%   Δp: 12.281%
90.0%   Δp:  1.502%
75.0%   Δp:  0.002%
Median  Δp: -0.000%
25.0%   Δp: -0.022%
10.0%   Δp: -4.132%
 5.0%   Δp: -20.603%
 1.0%   Δp: -80.709%
 0.1%   Δp: -99.912%
Minimum Δp: -100.000%
RMS Δp    : 17.650 ± 0.098 %
Same top p: 55.808 ± 0.129 %