llama_model_loader: loaded meta data with 49 key-value pairs and 720 tensors from gemma-4-E4B-it-WIP/gemma-4-E4B-it-Q6_K.gguf (version GGUF V3 (latest))
llama_model_loader: - type  f32:  339 tensors
llama_model_loader: - type  f16:   32 tensors
llama_model_loader: - type q5_1:    8 tensors
llama_model_loader: - type q8_0:   91 tensors
llama_model_loader: - type q5_K:    2 tensors
llama_model_loader: - type q6_K:  248 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q6_K
print_info: file size   = 5.69 GiB (6.50 BPW) 

====== Perplexity statistics ======
Mean PPL(Q)                   :  57.395025 ±   0.709393
Mean PPL(base)                :  52.693235 ±   0.610088
Cor(ln(PPL(Q)), ln(PPL(base))):  98.51%
Mean ln(PPL(Q)/PPL(base))     :   0.085471 ±   0.002205
Mean PPL(Q)/PPL(base)         :   1.089229 ±   0.002402
Mean PPL(Q)-PPL(base)         :   4.701791 ±   0.150722

====== KL divergence statistics ======
Mean    KLD:   0.006872 ±   0.000100
Maximum KLD:   3.890319
99.9%   KLD:   0.332454
99.0%   KLD:   0.064667
95.0%   KLD:   0.022811
90.0%   KLD:   0.013986
Median  KLD:   0.002055
10.0%   KLD:   0.000015
 5.0%   KLD:   0.000002
 1.0%   KLD:  -0.000000
 0.1%   KLD:  -0.000002
Minimum KLD:  -0.000105

====== Token probability statistics ======
Mean    Δp: -0.077 ± 0.006 %
Maximum Δp: 75.114%
99.9%   Δp: 15.320%
99.0%   Δp:  6.360%
95.0%   Δp:  2.144%
90.0%   Δp:  0.862%
75.0%   Δp:  0.043%
Median  Δp: -0.000%
25.0%   Δp: -0.097%
10.0%   Δp: -1.173%
 5.0%   Δp: -2.602%
 1.0%   Δp: -6.889%
 0.1%   Δp: -17.803%
Minimum Δp: -88.628%
RMS Δp    :  2.229 ± 0.037 %
Same top p: 96.672 ± 0.047 %