llama_model_loader: loaded meta data with 49 key-value pairs and 720 tensors from gemma-4-E4B-it-WIP/gemma-4-E4B-it-Q7_K.gguf (version GGUF V3 (latest))
llama_model_loader: - type  f32:  339 tensors
llama_model_loader: - type  f16:   52 tensors
llama_model_loader: - type q5_1:    1 tensors
llama_model_loader: - type q8_0:  274 tensors
llama_model_loader: - type q6_K:   54 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q8_0
print_info: file size   = 6.56 GiB (7.50 BPW) 

====== Perplexity statistics ======
Mean PPL(Q)                   :  57.909227 ±   0.717870
Mean PPL(base)                :  52.693235 ±   0.610088
Cor(ln(PPL(Q)), ln(PPL(base))):  98.55%
Mean ln(PPL(Q)/PPL(base))     :   0.094390 ±   0.002196
Mean PPL(Q)/PPL(base)         :   1.098988 ±   0.002413
Mean PPL(Q)-PPL(base)         :   5.215993 ±   0.155827

====== KL divergence statistics ======
Mean    KLD:   0.003237 ±   0.000086
Maximum KLD:   8.411139
99.9%   KLD:   0.152386
99.0%   KLD:   0.029975
95.0%   KLD:   0.010175
90.0%   KLD:   0.006270
Median  KLD:   0.000917
10.0%   KLD:   0.000005
 5.0%   KLD:   0.000001
 1.0%   KLD:  -0.000001
 0.1%   KLD:  -0.000006
Minimum KLD:  -0.000048

====== Token probability statistics ======
Mean    Δp: -0.062 ± 0.004 %
Maximum Δp: 74.921%
99.9%   Δp: 10.826%
99.0%   Δp:  4.042%
95.0%   Δp:  1.411%
90.0%   Δp:  0.565%
75.0%   Δp:  0.027%
Median  Δp: -0.000%
25.0%   Δp: -0.072%
10.0%   Δp: -0.796%
 5.0%   Δp: -1.764%
 1.0%   Δp: -4.686%
 0.1%   Δp: -11.661%
Minimum Δp: -87.893%
RMS Δp    :  1.534 ± 0.034 %
Same top p: 97.743 ± 0.039 %