====== Perplexity statistics ======
Mean PPL(Q)                   :  30.107026 ±   0.329175
Mean PPL(base)                :   9.042786 ±   0.071503
Cor(ln(PPL(Q)), ln(PPL(base))):  81.96%
Mean ln(PPL(Q)/PPL(base))     :   1.202791 ±   0.006352
Mean PPL(Q)/PPL(base)         :   3.329397 ±   0.021147
Mean PPL(Q)-PPL(base)         :  21.064240 ±   0.273651

====== KL divergence statistics ======
Mean    KLD:   1.009233 ±   0.003785
Maximum KLD:  22.312464
99.9%   KLD:  12.292943
99.0%   KLD:   7.613831
99.0%   KLD:   7.613831
Median  KLD:   0.574097
10.0%   KLD:   0.016605
 5.0%   KLD:   0.004251
 1.0%   KLD:   0.000354
Minimum KLD:  -0.000000

====== Token probability statistics ======
Mean    Δp: -7.022 ± 0.073 %
Maximum Δp: 99.488%
99.9%   Δp: 84.278%
99.0%   Δp: 60.487%
95.0%   Δp: 32.046%
90.0%   Δp: 16.402%
75.0%   Δp:  1.023%
Median  Δp: -0.418%
25.0%   Δp: -10.160%
10.0%   Δp: -43.946%
 5.0%   Δp: -72.034%
 1.0%   Δp: -98.340%
 0.1%   Δp: -99.994%
Minimum Δp: -100.000%
RMS Δp    : 28.800 ± 0.088 %
Same top p: 67.476 ± 0.122 %