llama_model_loader: loaded meta data with 49 key-value pairs and 720 tensors from gemma-4-E4B-it-WIP/gemma-4-E4B-it-Q8_0.gguf (version GGUF V3 (latest))
llama_model_loader: - type  f32:  339 tensors
llama_model_loader: - type  f16:   60 tensors
llama_model_loader: - type q8_0:  309 tensors
llama_model_loader: - type q6_K:   12 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q8_0
print_info: file size   = 7.44 GiB (8.50 BPW) 

====== Perplexity statistics ======
Mean PPL(Q)                   :  57.324375 ±   0.708968
Mean PPL(base)                :  52.693235 ±   0.610088
Cor(ln(PPL(Q)), ln(PPL(base))):  98.58%
Mean ln(PPL(Q)/PPL(base))     :   0.084239 ±   0.002164
Mean PPL(Q)/PPL(base)         :   1.087889 ±   0.002354
Mean PPL(Q)-PPL(base)         :   4.631141 ±   0.148444

====== KL divergence statistics ======
Mean    KLD:   0.001798 ±   0.000059
Maximum KLD:   4.473496
99.9%   KLD:   0.106441
99.0%   KLD:   0.015963
95.0%   KLD:   0.005257
90.0%   KLD:   0.003187
Median  KLD:   0.000441
10.0%   KLD:   0.000003
 5.0%   KLD:   0.000000
 1.0%   KLD:  -0.000001
 0.1%   KLD:  -0.000007
Minimum KLD:  -0.000047

====== Token probability statistics ======
Mean    Δp: -0.020 ± 0.003 %
Maximum Δp: 71.373%
99.9%   Δp:  8.424%
99.0%   Δp:  3.038%
95.0%   Δp:  1.051%
90.0%   Δp:  0.443%
75.0%   Δp:  0.026%
Median  Δp: -0.000%
25.0%   Δp: -0.037%
10.0%   Δp: -0.510%
 5.0%   Δp: -1.171%
 1.0%   Δp: -3.218%
 0.1%   Δp: -8.574%
Minimum Δp: -88.102%
RMS Δp    :  1.155 ± 0.038 %
Same top p: 98.371 ± 0.033 %