====== Perplexity statistics ======
Mean PPL(Q)                   :   8.602292 ±   0.060945
Mean PPL(base)                :   7.669212 ±   0.052592
Cor(ln(PPL(Q)), ln(PPL(base))):  97.95%
Mean ln(PPL(Q)/PPL(base))     :   0.114815 ±   0.001428
Mean PPL(Q)/PPL(base)         :   1.121666 ±   0.001602
Mean PPL(Q)-PPL(base)         :   0.933080 ±   0.014176

====== KL divergence statistics ======
Mean    KLD:   0.108536 ±   0.000393
Maximum KLD:   6.181332
99.9%   KLD:   1.618752
99.0%   KLD:   0.678996
99.0%   KLD:   0.678996
Median  KLD:   0.069372
10.0%   KLD:   0.001637
 5.0%   KLD:   0.000372
 1.0%   KLD:   0.000039
Minimum KLD:  -0.000000

====== Token probability statistics ======
Mean    Δp: -1.298 ± 0.024 %
Maximum Δp: 90.866%
99.9%   Δp: 43.591%
99.0%   Δp: 24.341%
95.0%   Δp: 11.986%
90.0%   Δp:  6.674%
75.0%   Δp:  0.897%
Median  Δp: -0.059%
25.0%   Δp: -3.073%
10.0%   Δp: -11.016%
 5.0%   Δp: -17.336%
 1.0%   Δp: -32.184%
 0.1%   Δp: -58.795%
Minimum Δp: -99.298%
RMS Δp    :  9.309 ± 0.040 %
Same top p: 83.878 ± 0.094 %