llama_model_loader: loaded meta data with 49 key-value pairs and 720 tensors from gemma-4-E4B-it-WIP/gemma-4-E4B-it-Q2_K.gguf (version GGUF V3 (latest))
llama_model_loader: - type  f32:  339 tensors
llama_model_loader: - type  f16:    1 tensors
llama_model_loader: - type q4_1:    7 tensors
llama_model_loader: - type q2_K:  112 tensors
llama_model_loader: - type q3_K:   36 tensors
llama_model_loader: - type q4_K:   15 tensors
llama_model_loader: - type q5_K:    2 tensors
llama_model_loader: - type iq2_xxs:   94 tensors
llama_model_loader: - type iq2_xs:   38 tensors
llama_model_loader: - type iq3_xxs:   38 tensors
llama_model_loader: - type iq4_xs:    2 tensors
llama_model_loader: - type iq1_m:   36 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = IQ2_S - 2.5 bpw
print_info: file size   = 2.19 GiB (2.50 BPW) 

====== Perplexity statistics ======
Mean PPL(Q)                   : 158.513459 ±   1.991659
Mean PPL(base)                :  52.693235 ±   0.610088
Cor(ln(PPL(Q)), ln(PPL(base))):  70.11%
Mean ln(PPL(Q)/PPL(base))     :   1.101352 ±   0.009377
Mean PPL(Q)/PPL(base)         :   3.008232 ±   0.028209
Mean PPL(Q)-PPL(base)         : 105.820225 ±   1.623290

====== KL divergence statistics ======
Mean    KLD:   2.371872 ±   0.006399
Maximum KLD:  30.074741
99.9%   KLD:  16.281898
99.0%   KLD:  10.876118
95.0%   KLD:   7.153347
90.0%   KLD:   5.582164
Median  KLD:   1.691821
10.0%   KLD:   0.054692
 5.0%   KLD:   0.010447
 1.0%   KLD:   0.000600
 0.1%   KLD:   0.000037
Minimum KLD:   0.000000

====== Token probability statistics ======
Mean    Δp: -13.185 ± 0.082 %
Maximum Δp: 99.835%
99.9%   Δp: 89.775%
99.0%   Δp: 59.588%
95.0%   Δp: 19.010%
90.0%   Δp:  4.926%
75.0%   Δp:  0.030%
Median  Δp: -0.543%
25.0%   Δp: -18.065%
10.0%   Δp: -66.818%
 5.0%   Δp: -88.176%
 1.0%   Δp: -99.296%
 0.1%   Δp: -99.987%
Minimum Δp: -100.000%
RMS Δp    : 33.940 ± 0.093 %
Same top p: 44.998 ± 0.130 %