llama_model_loader: loaded meta data with 49 key-value pairs and 720 tensors from gemma-4-E4B-it-WIP/gemma-4-E4B-it-Q4_K.gguf (version GGUF V3 (latest))
llama_model_loader: - type  f32:  339 tensors
llama_model_loader: - type  f16:    2 tensors
llama_model_loader: - type q4_1:    5 tensors
llama_model_loader: - type q5_1:   24 tensors
llama_model_loader: - type q8_0:    2 tensors
llama_model_loader: - type q3_K:    1 tensors
llama_model_loader: - type q4_K:   95 tensors
llama_model_loader: - type q5_K:  101 tensors
llama_model_loader: - type q6_K:    2 tensors
llama_model_loader: - type iq4_xs:  149 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q4_K - Medium
print_info: file size   = 3.94 GiB (4.50 BPW) 

====== Perplexity statistics ======
Mean PPL(Q)                   :  56.460364 ±   0.691942
Mean PPL(base)                :  52.693235 ±   0.610088
Cor(ln(PPL(Q)), ln(PPL(base))):  97.35%
Mean ln(PPL(Q)/PPL(base))     :   0.069052 ±   0.002823
Mean PPL(Q)/PPL(base)         :   1.071492 ±   0.003025
Mean PPL(Q)-PPL(base)         :   3.767129 ±   0.170411

====== KL divergence statistics ======
Mean    KLD:   0.095300 ±   0.000557
Maximum KLD:  15.853424
99.9%   KLD:   2.527996
99.0%   KLD:   0.863295
95.0%   KLD:   0.356812
90.0%   KLD:   0.229792
Median  KLD:   0.035697
10.0%   KLD:   0.000216
 5.0%   KLD:   0.000025
 1.0%   KLD:   0.000001
 0.1%   KLD:  -0.000001
Minimum KLD:  -0.000043

====== Token probability statistics ======
Mean    Δp: -0.421 ± 0.020 %
Maximum Δp: 99.633%
99.9%   Δp: 54.652%
99.0%   Δp: 23.879%
95.0%   Δp:  8.503%
90.0%   Δp:  3.458%
75.0%   Δp:  0.202%
Median  Δp: -0.000%
25.0%   Δp: -0.392%
10.0%   Δp: -5.060%
 5.0%   Δp: -11.030%
 1.0%   Δp: -29.030%
 0.1%   Δp: -62.118%
Minimum Δp: -98.611%
RMS Δp    :  7.866 ± 0.054 %
Same top p: 87.367 ± 0.087 %