llama_model_loader: loaded meta data with 49 key-value pairs and 720 tensors from gemma-4-E4B-it-WIP/gemma-4-E4B-it-Q5_K.gguf (version GGUF V3 (latest))
llama_model_loader: - type  f32:  339 tensors
llama_model_loader: - type  f16:   13 tensors
llama_model_loader: - type q5_1:   28 tensors
llama_model_loader: - type q8_0:    6 tensors
llama_model_loader: - type q4_K:    3 tensors
llama_model_loader: - type q5_K:  158 tensors
llama_model_loader: - type q6_K:  173 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q5_K - Medium
print_info: file size   = 4.81 GiB (5.50 BPW) 

====== Perplexity statistics ======
Mean PPL(Q)                   :  58.576787 ±   0.727545
Mean PPL(base)                :  52.693235 ±   0.610088
Cor(ln(PPL(Q)), ln(PPL(base))):  98.32%
Mean ln(PPL(Q)/PPL(base))     :   0.105851 ±   0.002352
Mean PPL(Q)/PPL(base)         :   1.111657 ±   0.002615
Mean PPL(Q)-PPL(base)         :   5.883553 ±   0.169353

====== KL divergence statistics ======
Mean    KLD:   0.021173 ±   0.000200
Maximum KLD:   8.555525
99.9%   KLD:   0.762584
99.0%   KLD:   0.206188
95.0%   KLD:   0.075586
90.0%   KLD:   0.047285
Median  KLD:   0.007055
10.0%   KLD:   0.000044
 5.0%   KLD:   0.000005
 1.0%   KLD:  -0.000000
 0.1%   KLD:  -0.000002
Minimum KLD:  -0.000057

====== Token probability statistics ======
Mean    Δp: -0.211 ± 0.010 %
Maximum Δp: 95.665%
99.9%   Δp: 29.084%
99.0%   Δp: 10.904%
95.0%   Δp:  3.685%
90.0%   Δp:  1.450%
75.0%   Δp:  0.063%
Median  Δp: -0.000%
25.0%   Δp: -0.216%
10.0%   Δp: -2.289%
 5.0%   Δp: -4.984%
 1.0%   Δp: -12.990%
 0.1%   Δp: -30.230%
Minimum Δp: -92.949%
RMS Δp    :  3.844 ± 0.042 %
Same top p: 94.144 ± 0.061 %